Adds the missing `graph_description_keccak.test.cpp` for keccak in
`boomerang_value_detection` module.

Updates the commit hash in blake files included in the blake audit scope
document.

## Summary
- Enable `BB_ASSERT` macros in WASM builds for improved debugging and
error detection
- Add `BB_ASSERT_NO_WASM` family for performance-critical asserts that
are too expensive to run in WASM
- Convert two high-frequency asserts to NO_WASM variants (RefArray
iterator at 159M hits, Polynomial::set_if_valid_index at 13.5M hits)
- Convert `throw_or_abort` in recursion_constraint and
public_input_component back to `BB_ASSERT` now that WASM asserts are
enabled

Closes AztecProtocol/barretenberg#1200

## Benchmark Results

### ChonkBench (WASM Full/2)
| Configuration | Time | Regression |
|--------------|------|------------|
| Baseline (asserts disabled) | 21,872 ms | - |
| All asserts enabled | 22,480 ms | +2.8% |
| With NO_WASM on hot asserts | 22,195 ms | +1.5% |

### AVM Bulk Test
| Configuration | Proving Time | Bulk Test Time |
|--------------|--------------|----------------|
| Baseline (asserts disabled) | 29,303 ms | 29,889 ms |
| With asserts enabled | 27,315 ms | 27,861 ms |

No significant regression observed in AVM proving.

The 1.5% regression in ChonkBench is an acceptable tradeoff for the
improved debuggability and error detection that asserts provide in WASM
builds.

## Test Plan
- [x] WASM ChonkBench Full/2 and Full/5 benchmarks pass
- [x] AVM bulk proving test passes (`avm_bulk.test.ts`)
- [x] Native builds unaffected (NO_WASM macros expand to regular
asserts)

This removes some unnecessary wrappers around the inputs to some
blackbox functions