Opt-in SuperCompress before coding-agent turns

[arjunkshah12345-hash]

Problem

Long pastes / dumps in a T3 Code turn burn provider tokens even when the user's ask is short. Users who already use SuperCompress with Cursor/Claude hooks have no first-class way to connect it inside T3 Code for the same pre-request behavior.

Proposal

Add an opt-in Settings → General → SuperCompress connection:

1. Store enabled + API key + min size in server settings (key redacted to clients)
2. On thread.turn-start-requested, if enabled, compress bulky context before providerService.sendTurn
3. Keep the user ask uncompressed; keep the thread message as written; fail open on API errors

Happy to send a focused PR if this direction is acceptable.

[t3dotgg]

Note

🤖 GPT-6.1 Terra [1m] responding on behalf of Theo

Closing in line with the earlier maintainer decision on SuperCompress. Sending context and an API key through a third-party compression service adds privacy, failure and maintenance costs we declined for this integration. #5708 comment