bench(python): add pytest-benchmark suite with evaluation and operator benchmarks

[aepfli]

Summary

• Replace manual time.time() benchmark with proper pytest-benchmark suite
• Add 19 benchmark scenarios covering evaluation, custom operators, state management, and concurrency
• Add optional comparison against pure-Python JSON Logic library
• Add pytest-benchmark>=4.0 to dev dependencies

Benchmark Scenarios

Category	Benchmarks	What they measure
Evaluation	bool simple, targeting match/no-match, string, int, float, object, large context	Flag evaluation across types and context sizes
Custom Operators	fractional, semver, starts_with, ends_with	Individual operator performance
State Management	5/50/200 flags, no-change re-apply	update_state() scaling
Concurrent	4-thread evaluation	Thread safety and contention
Comparison	vs pure-Python json-logic (optional)	Native PyO3 vs alternative

How to run

cd python
uv sync --group dev
maturin develop
pytest benchmarks/ --benchmark-only --benchmark-disable-gc -v

Initial Results

• Boolean simple: ~714ns/call (~1.4M ops/sec)
• Targeting match: ~1.7us/call (~594K ops/sec)
• Custom operators: ~1.3-4.6us/call
• State updates: ~24us (5 flags) to ~1ms (200 flags)

Test plan

• All 18 benchmarks pass (1 skipped - optional comparison library)
• maturin develop + pytest benchmarks/ succeeds
• Verify results are stable across runs

Closes #62

🤖 Generated with Claude Code