feat(google-chat,ms-teams): emit classifier facets on chat links

[KrisBraun]

Summary

Google Chat and Microsoft Teams links now carry classifier facets ({format, automation, reach}), completing the sweep that gave Slack (#381) and the other messaging connectors this signal. The facets let the platform tell real people's conversations apart from bot traffic without content inspection.

• Direct and group chats are facetted {format: chat, reach: direct}, with automation judged over the whole message batch: only an all-bot conversation is automated, and a single human-shaped message keeps it human. This is deliberately fail-open — a muteable automated verdict that hides a real person is the one failure classification must never produce.
• Named-space / channel threads are facetted reach: list, judged from the thread's parent message, with long posts reading as message rather than chat (same 1000-char rule as Slack).
• Bot detection uses each provider's own signal: Google Chat's explicit sender.type (HUMAN/BOT); Teams' from.application (or a message with no human sender). Teams system event messages are excluded from the judgement, mirroring their exclusion from the synced notes.

Google Chat's space type isn't on the message resource, so the space/dm distinction is passed into transformChatThread from the sync path, which always knows which kind of space it is walking.

Test plan

• New unit coverage for googleChatThreadFacets / googleChatDmFacets / teamsChannelFacets / teamsDmFacets, including the fail-open cases (mixed batches, empty batches, missing sender types) and the stripped-HTML length rule.
• Transform-level assertions that both connectors stamp facets on the links they build (space vs dm reach, whole-batch DM judgement).
• google-chat gains its vitest setup (config + script + devDep); existing ms-teams and slack suites still pass.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FHoQc2TdA62Wp4BwUxvR6P