run_with_progress can livelock on closed progress channel

[ndenev]

Summary

run_with_progress can livelock when the broadcast progress channel is closed.

Affected code

• src/progress.rs (run_with_progress)

Root cause

run_with_progress uses:

• tokio::select! { biased; ... }
• one branch awaiting progress_rx.recv()
• one branch awaiting the main future

When the sender side drops, progress_rx.recv() returns Err(Closed) immediately and keeps being ready forever. With biased;, that branch is always preferred, and because closed errors are ignored, the loop can spin and starve completion.

Impact

• Potential hang during interactive startup/context switch operations using run_with_progress.
• Potential CPU spin in closed-channel edge cases.

Proposed fix

1. Handle recv errors explicitly:
   • Err(RecvError::Closed): disable progress polling (or break) and await the main future directly.
   • Err(RecvError::Lagged(_)): continue.
2. Optionally remove biased; to reduce starvation risk, but explicit Closed handling is still required.

Tests to add

1. Closed receiver + pending future: function should still return when future completes (no hang).
2. Lagged updates do not abort operation.
3. Regression test with timeout to guard against infinite loop.