feat(slack): outbound file sharing — extract code blocks and upload to Slack (#282)

[pavshulin]

Summary

• Extract fenced code blocks (CSV, JSON, HTML, YAML, SQL, Python, JS, TS) from agent text responses and upload them as native Slack file attachments via the V2 Upload API
• Channel-agnostic extraction in message_router.py, Slack-specific upload in slack_adapter.py / slack_service.py
• files:write OAuth scope added — requires Slack app reinstall to workspace
• 27 unit tests covering extraction logic, language mapping, limits, and edge cases

How it works

Agent responds with:     "Here's the data: ```csv\nname,revenue\nAcme,150000\n...```"
Router extracts:         response_1.csv (the CSV content)
Slack receives:          File attachment + clean text "(see attached: response_1.csv)"

Security

No change to agent tool restrictions (WebSearch, WebFetch only for Slack users). Extraction operates on the agent's text response — no file system access, no Write/Bash tools granted.

Limits

Limit	Value
Min block size	100 chars
Max files per response	5
Max per-file size	500 KB
Max total size	2 MB

Deployment note

After deploying, reinstall the Slack app to the workspace to pick up the files:write scope (same process as files:read for #222).

Test plan

• 27 unit tests pass (tests/unit/test_slack_outbound_files.py)
• Integration: Slack user asks agent for CSV data → receives downloadable file
• Verify CSV opens correctly in Excel/Numbers
• Verify small code blocks (<100 chars) stay inline
• Verify non-data languages (bash, rust) stay inline
• Verify text message delivered even if upload fails

Closes #282

🤖 Generated with Claude Code

[vybe]

LGTM — clean channel-agnostic design, solid V2 Upload API implementation, good test coverage (27 tests). Ship it.