Hydra fixes PersistentQueue deadlock and adds operator traces

What happened

Hydra merged changes to its PersistentQueue handling after a deadlock condition could leave the queue’s head item stuck. When the internal queue was full, that condition could block all message broadcasts indefinitely and produce no log output.

The update changes queue removal to pop by index rather than rely on the false equality result described in the source record. It also adds regression tests for the failure mode.

Why it matters

For Hydra operators, the practical concern was a silent broadcast freeze at queue capacity. The new PersistentQueueFull, LoadFailed, and DeleteFailed traces are intended to make queue pressure and related failures easier to identify.

The change also makes delete failures non-fatal, which should make that failure path more recoverable according to the record. Operators using affected builds can watch the new traces as an early signal of queue pressure.

What to watch next

The next useful receipt is a Hydra node release that includes these merged commits, followed by operator reports showing how the new traces behave in use. The supplied record does not identify a released version or describe deployment impact.