Summary

There are 192 cores in our CI machines, the original threads are counted as following:

• Rayon workers: no limit, 192 by default
• Tokio workers: no limit, 192 by default
• Tokio blocking workers: limit to 256
  The total threads sum up to 640, exceeding the 500 max threads limitation of valgrind.

This pr limit rayon workers to 1, tokio workers to 8, and tokio blocking worker to 8. Valgrind is the single thread simulator, so less threads means faster CI and more stable bench results. But I'm afraid rspack will go to deadlock if we set a too small number of tokio workers.

This pr also switch @codspeed/actions to the official one.

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).