fix: preserve entity evidence fields in source event CSV

[Pigbibi]

Summary

• Preserve entity_match_type, match_evidence, and relationship_type in the RSS/source extractor's CSV output, alongside the existing eight columns.
• Keep all existing extraction defaults: unverified, empty evidence, and unverified. Source confidence does not establish a company relationship.
• Add CSV round-trip coverage and regressions for generic Strategy, cybersecurity, crypto-assets, and incidental WhatsApp mentions.
• Update the existing English and Chinese source-setup notes to describe this stage and the protected-branch human handoff.

Scope

Exactly four files: the source extractor, its test module, and the two source-setup notes. The official importer from #60, aliases, workflows, permissions, QAR acceptance rules, and live data are unchanged.

This PR does not generate real data or restore company-level evidence. No collection, model call, workflow dispatch, live-data publication, or deployment was performed. Authentic company evidence and any data PR require a separate reviewed step.

Validation

• Before the implementation: 5 new regression cases failed because the serialized CSV lacked the three fields; 5 existing extractor tests passed.
• After the implementation: 32 focused extractor/importer/handoff/tracker tests passed.
• Full suite: 140 passed.
• Independent review: 26 extractor/importer tests passed.
• Ruff on the changed Python files, compileall over src/scripts/tests, and whitespace checks passed.

Base verified before publication: 94c9cd0d36fa07303236301f1eb352db1a0c7e4f. Keep the normal required test check and protected-branch review process; this is a draft PR.