Speedup of coherence

Achieved using:
- chunks of combinations (reduces memory storage)
- component indicies (avoid mask and unues combinations)
- precomputation of constants
- possibility to compute in np.float32 or np.float64
1 job for f/coherence_speedup in 1 minute and 10 seconds (queued for 1 second)
latest
Status Name Job ID Coverage
  Test
passed test_pyconturb #31710
python

00:01:10