Long-running tasks give no feedback on whether they're still processing or stuck

When I submit a long task prompt in a workspace, the UI sometimes shows a loading indicator indefinitely, with no way to tell whether the agent is still working or has stalled.

I'd like proper feedback in both cases:

- If it's still processing, show that it's actively working.
- If it has stopped/stalled, let me know clearly.

A loading indicator with no further feedback either way is not acceptable.

I love this project and would really appreciate seeing this improved.

No files or tests are named; start by tracing the workspace prompt submission flow and the loading state used while an agent runs. Identify the existing signals for active work and stalled or stopped tasks, then verify that the UI clearly distinguishes both states for long-running prompts.

Written by the indexing model from the issue text.