extracted spectrum S and true spectrum S = R * M  where R is
resolution matrix is preserved exactly after the coadd

and fix the bug with masked pixels uncovered by the test

…nction

There was some divergence there already i.e.
gradivar=1./gradvar vs gradivar = (gradvar>0)/(gradvar+(gradvar==0))

The test case was also incorrect

When resolution matrices were added, the non-masked ivars were used.
(this is kinda irrelevant with the oncoming fix, but important first step)

Now it passes a similar dedicated test

Also add a generic test of coadd_cameras validating the coadded matrix
The currently added test cannot be fully enabled, (it was enabled with the threshold instead)
because of what happens with edge pixels and how to interpret resolution matrix there

of the flux + resolution matrix. It now only exclude the pixels
that are next to the arm edges

if mask is None
instead set to the mask to zero if there is no one. this way we
need just one code-path

in all exposures
Also move the resolution_weight into sepearate function

Make sure that if the combined ivar is zero from masked pixels we still
get preserve the ivar and resolution matrix

leading to large gradient
Add a test of cosmics

i.e. the only way to get mask=0 in a pixel must be to have
at least one mask=0 pixel in the stack

…resol_fix

…resol_fix

…resol_fix

add test for coadd_cameras bits

one when the comismic maskign would mask completely unrelated
pixels when one of the spectra in the set is fully masked

Add camera option to mask_cosmics for printing
Also put the check for the minimum number of spectra needed for cosmic masking
inside the function to avoid having it in two places

(motivated by desire of having all the lines of coadd()/coadd_cameras()
test-covered)

Move the chi2 threshold into a function
also get rid of harcoded value of 3 for min_cosmic rejection

deviate clearly more than prescribed by variance vector.
In that case I ignore the chi^2 stats and just try to
empirically estimate the threshold. It will never work
particularly well, but it is better than doing what we
are doing

make a much cleaner interface for mask_cosmics as it does not need to get
all the spectra + indices + mask
Now it will just accept the spectra that needs masking and it will return
the boolean mask of cosmics.
That will also make it easier to store that information in the future

…resol_fix