Adjust default thread count on Apple Silicon systems

[robertknight]

Testing this whisper demo on an M-series MacBook Pro with 5 performance cores and 6 efficiency cores I found that the performance improved when increasing RTEN_NUM_THREADS up to 5 and then started to drop.

RTen currently defaults to using a number of threads that matches the number of physical cores without consideration of whether the cores are all the same or a mix of performance/efficiency. I have also been doing most testing on systems with 4 or fewer physical cores, so there might be contention issues that come up into play with more.