fix: don't double-wrap WebRTCError raised inside connect()

[AdirAmsalem]

Summary

Errors surfaced through realtime.connect() were being reported twice and with a redundant nested cause. When a WebRTCError was raised from inside connect() (e.g. a server capacity rejection, or an initial image/prompt failure), the outer except Exception handler re-wrapped it into another WebRTCError and emitted it through the on_error callback a second time — once from the original raise site, once from the except block.

Callers now see a single, unwrapped error; on_error fires once per failure. Follow-up to #47, addressing review feedback.

Test plan

• New regression test asserts no nested cause and no duplicate on_error invocation
• Full realtime unit suite still passes (38/38)

---

Note

Low Risk
Small, localized error-handling change with added unit coverage; main risk is subtly altering callback firing order/behavior on edge-case connection failures.

Overview
Prevents WebRTCConnection.connect() from double-wrapping a WebRTCError (and from emitting on_error twice) by adding a per-connect() _on_error_fired guard and a dedicated except WebRTCError path that re-raises as-is.

Adds regression tests covering both a server-error/_handle_error race and direct WebRTCError raises (e.g., ack timeouts), asserting no nested cause and exactly one on_error callback invocation.

^{Reviewed by Cursor Bugbot for commit a40c2ad. 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.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit e9ab59b. Configure here.}