BUG: run with no eligible worker waits silently forever (no fail-fast on permanent dead-end)

[addadi]

Bug: run with no eligible worker waits silently forever (no fail-fast on permanent dead-end)

Summary

When selectWorker returns null, the engine treats every case identically: retry with backoff. It cannot distinguish:

• transient backpressure — matching workers exist but are saturated → retrying is correct;
• permanent dead-end — no active worker matches the tags at all (wrong tag, worker dead/drained, never registered) → retrying loops forever.

In the permanent case the run sits in running state indefinitely with no terminal failure, no surfaced error, and repeated retry log spam.

Root cause

src/dispatch.zig:37 selectWorker returns null for both conditions; the engine's retry loop has no way to ask "does any active worker ever match these tags?".

Reproduction

1. Submit a run whose step requires worker_tags: ["nonexistent-tag"].
2. Run never terminates; steps retry forever; /runs/{id} shows running with repeated attempts.

Expected behavior

A hasActiveWorkerForTags(tags)-style check (active status + tag intersection, ignoring load) distinguishes the two cases:

• matching active worker exists but is saturated → keep retrying (backpressure);
• no matching active worker → fail the run with a clear error ("no worker can serve node X (tags [...])").

This also matters for saturation policy: a single saturated worker with max_concurrent=1 should not be mistaken for a dead-end.