Thread wakening may be bottom neck for large core systems

Result: boost from `3.6~8token/s` to `4.2~4.3token/s` on SG2042.

Analyze: I added tracy to trace execution in details. I observed that worker wakening is problematic. **There are some workers picked up task after the completion of other workers.**

That means the execution time is as twice as the expected sometimes. I guess that it is caused by `thread::yield()`, which switched out the busy waiting loop.

I am thinking about to make a higher/busier performance poll, while being able to switch to an idle state when waiting for user input.

Start with the Tracy traces described in the issue and investigate the worker-wakening path around thread::yield(). Reproduce the delayed task pickup on SG2042, then measure whether a busier polling strategy improves throughput without preventing the workers from becoming idle while waiting for user input.

Written by the indexing model from the issue text.