acdbot: introduce acdt breakout architecture

[wolovim]

• problem: capturing breakout rooms in acdt is a manual process that no one has bandwidth for
• solution: continue the EL breakout on the main acdt call and create a new recurring zoom meeting for the CL breakout that participants will switch to after the main call concludes. asset pipeline updated for this new reality.
• impl details:
  • acdt series mapping now includes "breakout_meeting_ids"
  • looks for the CL breakout each time an acdt call is processed in the pipeline and reacts accordingly: publishes meeting link in github comment, pulls zoom assets, publishes to youtube
  • in the interest of maintenance simplicity, removed a feature that composed the intro video onto acdt calls

[dionysuz-bot]

Blocking issue:

• .github/ACDbot/scripts/asset_pipeline/generate_manifest.py:31 hard-codes only the cl breakout resources even though the rest of the implementation models breakout labels as data via breakout_meeting_ids and writes transcript_{label}.vtt, chat_{label}.txt, and tldr_{label}.json. That means any future configured breakout label will download and summarize correctly, but then disappear from manifest.json, so downstream consumers will not see the artifacts. Please derive breakout manifest entries dynamically, either by scanning transcript_*.vtt / chat_*.txt / tldr_*.json with the corrected transcript excluded, or by iterating the configured breakout labels and constructing the expected filenames. This keeps the domain boundary data-driven instead of requiring code edits for each breakout label.

Validation run: uv run pytest from .github/ACDbot passed.