feat: add regex-backed detection benchmark strategies

[binaryaaron]

Summary

Adds benchmark-only regex/rule detection strategies for structured secrets and identifiers. The goal is to measure when deterministic detection can reduce LLM work while preserving anonymization safety.

Stack

• Base: feat: anonymizer measurement instrumentation and benchmark tooling #177 measurement instrumentation and benchmark infrastructure
• Previous: feat: add local structured substitute benchmark strategy #184 local structured substitute replacement-map strategy
• This PR: regex-backed detection benchmark strategies

What changed

• Add high-confidence rule detection for structured tokens such as API keys, credentials, cookies, URLs, identifiers, and shell-history secrets.
• Add benchmark strategy variants for rules-only, guardrail, filter-guardrail, covered-label routing, and native router probes.
• Add route/source-count analysis fields, staged probe summaries, signature-delta tooling, and regression coverage for regex-backed paths.
• Document the strategy set as benchmark probes, not public Anonymizer defaults.

Validation

• uv run --frozen ruff format ...
• uv run --frozen ruff check ...
• uv run pytest tests/engine/test_detection_rules.py tests/engine/test_structured_substitute.py tests/test_measurement.py tests/tools/test_benchmark_output_analysis.py tests/tools/test_compare_strategy_pairs.py tests/tools/test_detection_strategies.py tests/tools/test_extract_signature_deltas.py tests/tools/test_measurement_tools.py tests/tools/test_replacement_strategies.py tests/tools/test_replay_replacement_strategies.py tests/tools/test_screen_strategy_comparisons.py tests/tools/test_staged_detection_output_analysis.py tests/tools/test_staged_detection_probe.py -q
• git diff --check

Focused suite result: 256 passed, with existing DataDesigner provider deprecation warnings.