[Fix] Live task pages stall after initial connection failures

[roomote-roomote]

​Opened on behalf of @mrubens. View the task or mention @roomote-roomote for follow-up asks.

What changed

Initial live-task connection failures now retry in the background every 10 seconds after the bounded connection attempts are exhausted. Once a live connection has been established, the existing 60-second background recovery cadence is unchanged. The retry scheduler accepts a phase-aware interval and safely stops scheduling after cleanup.

Why this change was made

Recent replay evidence showed transient status-0 sandbox/setup requests leaving users on a "Could not reach the host" state that was abandoned before the visible retry action was used. The existing reconnect fix covered established connections, but initial-connection recovery still waited too long for another automatic attempt.

Impact

Users who open a task while its sandbox endpoint is temporarily unreachable get another automatic recovery attempt within the same interaction window, without changing manual Retry/Reconnect behavior or increasing the cadence for already-established connections. Focused client recovery tests, the related SandboxProvider and ErrorFallback suites, web fast type checks, targeted lint, formatting, and diff checks pass.

[roomote-community]

No code issues found. See task

_{Reviewed aa4d9e3}