test(perf): add Criterion microbenchmarks and expanded wrk scenarios

[ZhuchkaTriplesix]

Context

• perf-test/ covers hello-world wrk only; optional gate in tests/test_perf_hello_bench.py.
• No benches/ / Criterion in the repo; backlog perf: reduce lock contention on route match (RWLock or route snapshot) #4 suggested microbenches.
• Planned perf work (chore: Cursor rule — feature branches + PRs per issue #30–feat: JWT from Cookie header (issue #7) #33, feat: body_schema for OpenAPI request body without Pydantic (issue #9) #42–Dev #45) needs a regression baseline.

Problem

Optimizations cannot be proven or guarded without repeatable microbenchmarks and expanded scenario coverage.

Proposed change

1. Add Criterion benches (opt-in, not required in default CI) for:
   • match_route_compiled (static vs param route)
   • map_handler_return (str / bytes)
   • json_to_py (before/after direct conversion)
2. Extend perf-test with wrk scenarios:
   • plain text GET
   • JSON POST body + JSON response
   • JWT-protected GET
   • CORS-enabled app
   • route with one dependency
3. Document how to run locally; optional CI smoke with threshold (low connection count).

Files

• new benches/, Cargo.toml, perf-test/, perf-test/README.md
• optional workflow job (manual / scheduled)

Acceptance criteria

• cargo bench runs selected microbenches when configured.
• perf-test documents all scenarios and env vars (OXYROUTE_BENCH=1, etc.).
• README explains baseline collection before perf PRs.

Expected impact

Evidence-based perf work; regression detection for hot-path changes.

[ZhuchkaTriplesix]

Shipped in #134 (merged to dev).