Investigate runtime warnings from invalid `var` values in CPU correlator

[BSchilperoort]

With files from https://www.bonvinlab.org/protein-detective/em_prepare.html I am getting warnings now

powerfit emd_33292-7xm9-B-3.48.cheated.mrc 3.3 7xm9.B-reoriented.gemmi.pdb  -a 20  -l -d run-KsgA-py3f2
Target file read from: /home/stefanv/git/protein-detective/protein-detective/docs/emd_33292-7xm9-B-3.48.cheated.mrc
Target resolution: 3.30
Initial shape of density: 99 71 61
Shape after trimming: 94 71 59
Shape after extending: 96 72 60
Template file read from: /home/stefanv/git/protein-detective/protein-detective/docs/7xm9.B-reoriented.gemmi.pdb
Reading in rotations.
Requested rotational sampling density: 20.00
Real rotational sampling density: 20.83
Requested number of processors: 1
Starting search
/home/stefanv/git/protein-detective/powerfit/src/powerfit_em/correlators/cpu.py:155: RuntimeWarning: invalid value encountered in sqrt
  self.lcc_scan = np.where(self.vars.lcc_mask, self.vars.gcc / np.sqrt(var), 0.0)
/home/stefanv/git/protein-detective/powerfit/src/powerfit_em/correlators/cpu.py:155: RuntimeWarning: divide by zero encountered in divide
  self.lcc_scan = np.where(self.vars.lcc_mask, self.vars.gcc / np.sqrt(var), 0.0)
/home/stefanv/git/protein-detective/powerfit/src/powerfit_em/correlators/cpu.py:155: RuntimeWarning: invalid value encountered in divide
  self.lcc_scan = np.where(self.vars.lcc_mask, self.vars.gcc / np.sqrt(var), 0.0)
Processing rotations 100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 648/648  [ 0:00:16 < 0:00:00 , 39 rot/s ]
Time for search: 0m 19s
Analyzing results
Writing solutions to file.
Writing PDBs to file.
Total time: 0m 20s

The master branch does not show these errors, either solve or silence the warnings.

Originally posted by @sverhoeven in #66 (review)

[BSchilperoort]

Note that these runtime warnings started showing once we switched to computing the LCC with numpy instead of cython code. Cython code will not show these warnings, but the same NaN values will result from the computation.