[Enh] speed up sampling with emcee

[MuellerSeb]

The EnsembleSampler of emcee has an option vectorize, to state, that the log probability function accepts vectors: https://emcee.readthedocs.io/en/v3.1.4/user/sampler/?highlight=vectorize

Since this is the case for our implementations in the covariance model class, we should use it.

I could speed up the setup of a SRF class by factor 5 to 6:

import timeit
import gstools as gs
m = gs.Matern(dim=3)
%timeit gs.SRF(m)

• currently:

  1.23 s ± 89.8 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

• with vectorize=True

  231 ms ± 13 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

For models without an analytical spectral density, this is significant:

import timeit
import gstools as gs
m = gs.Stable(dim=3)
%timeit gs.SRF(m)

• currently:

  3.62 s ± 42.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

• with vectorize=True

  449 ms ± 13.2 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)