Reduce memory allocations by using bitset instead of sync.map

[djeebus]

AI results via benchtest:

● Benchmark Results                                                                                                                                                               
                                                                                                                                                                                  
  Performance Comparison (sync.Map → bitset)                                                                                                                                      
  ┌─────────────────────┬────────────────┬──────────────┬──────────────┐                                                                                                          
  │      Operation      │ sync.Map (old) │ bitset (new) │   Speedup    │                                                                                                          
  ├─────────────────────┼────────────────┼──────────────┼──────────────┤                                                                                                          
  │ SetIsCached         │ 374.70 ns/op   │ 20.28 ns/op  │ 18.5x faster │                                                                                                          
  ├─────────────────────┼────────────────┼──────────────┼──────────────┤                                                                                                          
  │ IsCached_Hit        │ 42.09 ns/op    │ 25.07 ns/op  │ 1.7x faster  │                                                                                                          
  ├─────────────────────┼────────────────┼──────────────┼──────────────┤                                                                                                          
  │ IsCached_Miss       │ 63.02 ns/op    │ 11.46 ns/op  │ 5.5x faster  │                                                                                                          
  ├─────────────────────┼────────────────┼──────────────┼──────────────┤                                                                                                          
  │ WriteAt_SingleBlock │ 707.1 ns/op    │ 676.2 ns/op  │ 1.05x faster │                                                                                                          
  ├─────────────────────┼────────────────┼──────────────┼──────────────┤                                                                                                          
  │ WriteAt_MultiBlock  │ 2244 ns/op     │ 1674 ns/op   │ 1.3x faster  │                                                                                                          
  └─────────────────────┴────────────────┴──────────────┴──────────────┘                                                                                                          
  Memory Allocations                                                                                                                                                              
  ┌─────────────────────┬────────────────────┬──────────────────┬───────────┐                                                                                                     
  │      Operation      │   sync.Map (old)   │   bitset (new)   │ Reduction │                                                                                                     
  ├─────────────────────┼────────────────────┼──────────────────┼───────────┤                                                                                                     
  │ SetIsCached         │ 56 B/op, 1 alloc   │ 0 B/op, 0 allocs │ 100%      │                                                                                                     
  ├─────────────────────┼────────────────────┼──────────────────┼───────────┤                                                                                                     
  │ IsCached_Hit        │ 0 B/op             │ 0 B/op           │ -         │                                                                                                     
  ├─────────────────────┼────────────────────┼──────────────────┼───────────┤                                                                                                     
  │ IsCached_Miss       │ 0 B/op             │ 0 B/op           │ -         │                                                                                                     
  ├─────────────────────┼────────────────────┼──────────────────┼───────────┤                                                                                                     
  │ WriteAt_SingleBlock │ 56 B/op, 1 alloc   │ 0 B/op, 0 allocs │ 100%      │                                                                                                     
  ├─────────────────────┼────────────────────┼──────────────────┼───────────┤                                                                                                     
  │ WriteAt_MultiBlock  │ 224 B/op, 7 allocs │ 0 B/op, 0 allocs │ 100%      │                                                                                                     
  └─────────────────────┴────────────────────┴──────────────────┴───────────┘                                                                                                     
  Summary                                                                                                                                                                         
                                                                                                                                                                                  
  - SetIsCached is 18.5x faster with zero allocations (was 1 alloc per call)                                                                                                      
  - IsCached_Miss is 5.5x faster (bitset Test is much faster than sync.Map Load miss)                                                                                             
  - Multi-block writes eliminated 7 allocations per operation (one per block touched)                                                                                             
  - Overall geomean speedup: 66.56%                                                                                                                                               
                                                                                                                                                                                  
  The biggest win is eliminating the per-operation allocations, which directly reduces GC pressure. In the original profile, setIsCached was responsible for 0.91s of CPU time    
  with 92% in sync.Map.Store - this should now be near zero.   ```

[cursor]

PR Summary

Medium Risk
Changes the core cache dirty-tracking data structure and range-calculation logic, which could impact correctness of cache hits/exports under edge cases or concurrency despite added tests.

Overview
Replaces block-cache dirty tracking from per-offset sync.Map entries to a pre-sized bitset.BitSet with its own lock to reduce allocations and speed up isCached/setIsCached checks and iteration during diff export. Tightens range handling by guarding against negative offsets, zero-length operations, and out-of-bounds writes/reads so these cases no longer incorrectly mark blocks as cached, and adds targeted unit tests plus benchmarks to validate behavior and measure the performance change.

^{Written by Cursor Bugbot for commit 8134a74. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

There are 3 total unresolved issues (including 1 from previous review).

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6a120a755c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".