[SYCL][NATIVECPU] faster enqueues and async ops

[uwedolinsky]

Aims to improve kernel enqueues to:

1. make the enqueuing faster by reducing work of the main thread and doing more work in the launched threads, including waiting for dependent events
2. launch less threads per enqueue (max Maxthreads and potentially one for peeling)
3. do peeling by the launched thread (instead of the main thread)

The state struct is now constructed by the threads instead of being copied which also reduces the capturing overhead in the lambda (also removing #ifdef NATIVECPU_USE_OCK)
Most other operations including memory copies are now asynchronous except enqueueBufferMap