test: add benchmarks for opt-in utilities

[Fieldnote-Echo]

Summary

Benchmarks exist for clean() and walk() in tests/test_benchmark.py, but the three opt-in utilities (decode_evasion, detect_scripts, is_mixed_script) have no performance baselines.

What to do

Add two benchmark classes to tests/test_benchmark.py:

TestDecodeBenchmarks

• Single-layer URL decoding — decode_evasion("%48%65%6C%6C%6F")
• Triple-nested decoding — URL-encode 3 levels deep (default max_depth=5)
• Clean text (no-op) — decode_evasion("Hello world")

TestScriptDetectionBenchmarks

• Pure Latin — detect_scripts("Hello world")
• Mixed script (phishing-like) — text with Cyrillic + Latin
• CJK-heavy text — paragraph of CJK characters
• is_mixed_script on mixed input — verify it runs quickly

Follow the existing benchmark patterns in the file — use benchmark() for small inputs and benchmark.pedantic() for large ones (100KB+).

Files

• tests/test_benchmark.py — add new classes

Hints

• Look at existing benchmarks for style: TestCleanBenchmarks, TestWalkBenchmarks
• The opt-in imports: from navi_sanitize import decode_evasion, detect_scripts, is_mixed_script
• Run benchmarks with: uv run pytest tests/test_benchmark.py -v
• Regular test runs skip benchmarks: --benchmark-disable