[ENH] Optimization

[ryanhammonds]

Addresses a point in #193. This adds minor tweaks to speed up fitting. The most noticeable difference is when using fit_fooof_3d. 56% faster for a (100, 10, 99) array:

import numpy as np

from fooof import FOOOFGroup
from fooof.objs import fit_fooof_3d
from fooof.sim import gen_group_power_spectra

np.random.seed(0)

ap_params = [-2, 0]
pe_params = [[10, 5, 5], [25, 5, 5]]

freqs, powers = gen_group_power_spectra(10, (1, 50), ap_params, pe_params)

# Shape: (100, 10, 99)
powers_3d = np.array([powers] * 100)

fg = FOOOFGroup(verbose=False)

%%timeit
fgs = fit_fooof_3d(fg, freqs, powers_3d, n_jobs=4)

Before:

3.56 s ± 20.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

After:

2 s ± 19.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

[TomDonoghue]

Thanks @ryanhammonds - this all looks great!

I checked through the code, and everything looks good - some really nice little clean ups and optimizations here.
I quick added some small code doc / notes extensions, and extended the test for fit_fooof_3d a little because when I was checking I figured it could / should do a little more checking that shapes are conserved as expected.

Merging this in now!