perf(codegen): use a fast hasher for internal naming/comment maps

[th0114nd]

Closes #396.

What this does

CodeGenContext's internal bookkeeping (context.rs's naming/deconfliction maps, comments.rs's doc-comment lookup) used std::collections::HashMap, which defaults to SipHash. This data only ever comes from trusted, locally-generated .proto descriptors — never attacker-controlled input — so SipHash's hash-flooding resistance buys nothing here, and profiling showed it as the single largest sampled cost bucket on a large multi-thousand-file codegen run (see #396 for the numbers).

Swaps context.rs and comments.rs's internal maps to foldhash via a small hash.rs type alias (FxHashMap). foldhash is already resolved transitively via hashbrown, so this is a direct dependency on something already in the tree.

oneof.rs's maps were tried too but reverted: their types leak through public signatures in message.rs/view.rs/lazy_view.rs, too broad a blast radius for what should be a purely internal, output-invariant change. Left as std::HashMap.

One field (CodeGenContext::field_renames) stays std::collections::HashMap explicitly — it's populated from field_names::plan_field_names, which isn't touched by this change, so its type has to match.

Testing

• Output-invariant by construction: verified with a full hash-diff of a real ~2900-crate generation run, before and after — byte-for-byte identical.
• All existing buffa-codegen tests pass unmodified (cargo test -p buffa-codegen --lib).
• cargo fmt, cargo clippy --workspace --all-targets -- -D warnings, and cargo doc --all-features all clean.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[th0114nd]

Closing this — on reflection, a ~3-7% wall-clock improvement (verified via interleaved trials after an earlier noisy measurement) doesn't clearly justify separate review attention on its own, especially alongside #399's ~2x improvement, which deserves the focus. Not pursuing further for now; may revisit later if it's ever bundled with something else that needs the same review pass.

Thanks for reading this far regardless — no action needed.