Select text detectors before scanning

[sidmohan0]

Problem

Text scans run every enabled detector even when callers only want certain entity types. MCP filters findings afterward, so unwanted matches still consume time and memory. See DataFog/datafog-mcp#59.

Solution

• Add optional entities selection to text scans and skip unselected detectors before matching.
• Preserve existing results when selection is omitted, along with Unicode offsets, ENV/SQL boundaries, and locale/UUID activation rules.
• Reject invalid selections through the existing configuration-error contract.
• Expose the option in Python, Node, and WASM types and verify it across the installed packages. Structured scanning remains unchanged.

Validation: Rust format, workspace/all-features Clippy, and 122 tests pass (one existing benchmark ignored). Python installed-wheel/typing, Node package/TypeScript, and WASM browser/TypeScript checks pass. Node validation used host Node 25; CI should confirm the declared Node 24 environment.

Five fresh-process runs per case on Mac arm64/Python 3.11, using 10 MB synthetic log fixtures from #59:

Case	Published Core: median scan / peak RSS	Selected candidate: median scan / peak RSS
No selected entities	0.243 s / 178 MB	0.114 s / 61 MB
~1,000 selected entities per MB	0.246 s / 179 MB	0.125 s / 66 MB

The control's 271,710 unrequested findings disappear; the mixed case returns the 9,999 planted selected findings. Candidate scans with selection omitted retain the published counts and similar performance. These are Core-only measurements, not end-to-end MCP/model benchmarks.

Closes #46. Downstream MCP integration is tracked in DataFog/datafog-mcp#63. No package version or release is changed; MCP adoption requires a subsequent Core release.

Draft MCP implementation: DataFog/datafog-mcp#65.