fix: Close idle MCP sessions

[jirispilka]

MCP connections don't properly close when clients disconnect, keeping containers alive indefinitely.
Implemented automatic session cleanup using internal DELETE requests after 60s timeout (in const.py).
This timeout should match idle timeout of the Actor standby settings as this variable is not exposed. Yes, it is a bit inconvenient.

I'll propagate these changes to other templates after this is reviewed.

Test run (created multi-sessions) with list/tool

[JJetmar]

As I understand it, there is an implicit functionality of the python mcp package causing the "sessions" to to be still alive by actively polling their status (or something) via HTTP requests, and that is what keeps our stand-by Actors running forever.

This looks like an effective fix.

[jirispilka]

As I understand it, there is an implicit functionality of the python mcp package causing the "sessions" to to be still alive by actively polling their status (or something) via HTTP requests, and that is what keeps our stand-by Actors running forever.

This looks like an effective fix.

I still don’t fully understand the underlying issue.
I tried different settings for the MCP server — json_response (so the server works in request → response mode), stateless config, and keep_alive = 1 — but none of them worked.

After spending some time on it, I decided to patch it temporarily by killing sessions internally and will figure out a proper solution later.