feat(browser): support stored-session followups

[paulbettner]

Summary

• add --engine browser --followup <sessionId> support for stored ChatGPT/GPT browser sessions
• allow attachment uploads and upload fallback for browser follow-ups when inline context is too large
• harden browser follow-up recovery so post-submit disconnects resume without resending
• validate browser follow-up constraints earlier and tighten the CLI/docs around browser-vs-API follow-up semantics
• classify resolved directory/glob inputs correctly in browser mode and enforce generated bundle size limits

Details

This keeps browser follow-ups on the stored browser session metadata instead of Responses API ids.

The hardening pass focuses on failure modes that are easy to miss in manual testing:

• existing-Chrome and reopened-Chrome follow-ups now resume safely after post-submit observation failures
• browser follow-ups reject unsupported surfaces up front (resp_... ids, --followup-model, Gemini browser sessions, and legacy browser sessions missing runtime metadata)
• browser prompt assembly now classifies resolved files after glob/directory expansion, so archives discovered through directories/globs route to browser uploads instead of text reads
• generated browser bundles now honor maxFileSizeBytes

Testing

• pnpm test
• pnpm run check
• pnpm run build