#0: Use CV to wait for cq_reader in production mode. Remove enqueue_record_event for NB calls

[tt-asaigal]

Ticket

Link to Github Issue

Problem description

Issuing a record_event when making non-blocking calls leads to bad perf in async mode, since the Completion Queue Reader and Worker Threads are bound to the same CPU core.

Similarly, a while loop in finish (polling the Completion Queue Reader status) is bad, due to worker polling CQ reader and CQ reader polling device.

What's changed

Remove record_event calls, since they're not needed for functionality.

Have CQ reader poll device and send an interrupt to worker thread, using condition vars. No more double polling, and worker thread yields CPU while waiting.

Checklist

• Post commit CI passes
• Model regression CI testing passes (if applicable)
• New/Existing tests provide coverage for changes