[codex] capture XHS creator IDs for private research pipeline

[yukikojo]

What changed

• capture 24-character public XHS creator IDs before anonymous note persistence
• expose a default-off CLI/API capture switch used by the keyword runner
• write private ID and invalid-ID sidecars atomically with strict allowlists
• keep raw IDs out of ordinary storage, logs, and AI-facing artifacts

Why

The keyword pipeline needs a local ID bridge so qualified public accounts can be crawled by creator mode for a second research screening pass.

Validation

• focused pytest suite: 27 passed, 2 Windows-incompatible repository scans deselected
• git diff check passed
• live keyword smoke test with 重庆企业家 captured 20 private creator-ID records without raw IDs in public or AI artifacts