[Threadpool] Remove reserve threads

[mratsim]

This removes a significant threadpool perf pessimization introduced in #222

• reserve threads are removed.
• some perf was also lost because:
  • primitives.nim was imported and it needed globals for CPU features detection. Unsure how this exactly slowed a benchmark down though.
  • creating a task is inlined for spawn (but not for loops as those requires more initialization)
  • the conditional branch in sync was a debugging aid and is not needed

An additional slight optimization is avoiding syscalls in wake when we have sleepy workers that are not yet asleep. The change in epoch is sufficient to avoid their sleep.

Finally, the stealHalf and adaptative theft policy code has been removed as it was never finished.

[mratsim]

Benchmarks

While many are unaffected, some are showing dramatic improvements

Log-sum-exp (up to 58x speedup)

Yup 58x speedup

Before

after

Heat diffusion (7.11x faster)

Before

After

Depth-First Search (up to 2x speedup)

Before

After

Fibonacci (1.45x faster)

Before

After

Crypto bench (6x improvement)

Focus on batch elliptic addition / sum reduction of 65536 BLS12-381 points

Before

After

The pessimization code was slower than serial while after fix the code is 5.9x faster than serial