fix: surface server capacity error through connect()

[AdirAmsalem]

Summary

When the realtime server rejects a connection because it's at capacity, it sends a structured error message and closes the WebSocket. The Python SDK was swallowing that message and surfacing a generic "WebSocket disconnected" instead — indistinguishable from a real network drop. In subscribe mode, there was no pending operation to deliver the error at all, so callers hung for the full 5-minute connection timeout before seeing any failure.

With this change, connect() raises WebRTCError("Server at capacity. Please try again later.") promptly, so callers can distinguish capacity rejection from a network drop and back off / inform the user appropriately. Matches the JS SDK's behavior.

Test plan

• New unit tests cover the receive-loop race and the subscribe-mode no-pending-waits case
• Existing realtime unit suite still passes (37/37)

---

Note

Low Risk
Small, localized error-propagation change in the realtime WebRTC connection flow, with added unit coverage and no data/auth implications.

Overview
WebRTCConnection now tracks a first server-sent connection error (_connection_error) and uses it to fail connect() fast (including subscribe mode) instead of waiting for the full timeout.

The WebSocket receive-loop’s cleanup no longer overwrites structured server errors with a generic "WebSocket disconnected"; it resolves pending waits using the preserved server error when present. New unit tests cover the disconnect race and the immediate connect() failure behavior.

^{Reviewed by Cursor Bugbot for commit 65cfbc3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit 65cfbc3. Configure here.}