fix incorrect use of polyfit

[segasai]

Hi,

while investigating trace_shifts I noticed that there is a couple places in desispec that use np.polyfit weights assuming those weigths are inverse variances, while they should be inverse errors:

desispec/py/desispec/trace_shifts.py

Line 569 in bf216b2

c = np.polyfit(fy,fdx,deg,w=1/fex**2)

desispec/py/desispec/qproc/qfiberflat.py

Line 83 in bf216b2

c = np.polyfit(x[jj],tflux[i,jj]/mflux[jj],1,w=mflux[jj]**2*tivar[i,jj])

cf. docs
https://numpy.org/doc/stable/reference/generated/numpy.polyfit.html

The patch fixes that. (I am assuming that the patch should be pretty safe)

[julienguy]

Wow, good catch. I should never use those built-in minimization (without reading the doc carefully!).