Optimize message encoding buffer reuse

[bravo1goingdark]

Summary

• reuse a thread-local FlatBufferBuilder and BytesMut so message encoding no longer allocates new buffers for every payload
• adjust the encoder helpers to return Bytes views while updating frame writers to reuse the pooled buffers
• extend the production benchmark with an allocation-counting encoder test and add the required dev dependency

Testing

• cargo check (fails: inherits must not be specified in root profile release)

---

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

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".

[chatgpt-codex-connector]

Avoid defining a second global allocator in benchmarks

The new allocation-counting benchmark unconditionally declares #[global_allocator] static GLOBAL_ALLOCATOR: CountingAllocator = CountingAllocator;. The library already exposes a #[global_allocator] when the mimalloc feature (and thus the ultra/production feature sets) is enabled in src/lib.rs. Running cargo bench --features mimalloc will now fail to compile with “the #[global_allocator] has already been defined” because both the library and the benchmark provide one. Consider guarding the counting allocator behind a cfg that disables it when mimalloc (or any feature that sets a global allocator) is active, or reusing the existing allocator via a wrapper instead of redefining it.

Useful? React with 👍 / 👎.