HLA-1312: Numpy 2.0 scalar promotion issues in STWCS

[s-goldman]

Resolves HLA-1312

Closes #199

This PR addresses issues related to scalar promotion in numpy 2.0. The main issues are related to the IDCTAB refpix values being downgraded. More precisely, they are no longer automatically being upgraded from a float32 to a float64.

The important value that is now manually upgraded is the 'PSCALE'. Without this change, the 'PSCALE' is truncated (e.g. for ACS, WFC3, and WFPC2 IDCTABs) with a float32.

Several other values in the IDCTAB ('XREF', 'YREF', and 'THETA') are no longer automatically being upgraded, but these do not affect the results as, while they are technically floats, the values are integers.

Checklist for maintainers

• added entry in CHANGELOG.rst within the relevant release section
• updated or added relevant tests
• updated relevant documentation
• added relevant label(s)

[mdlpstsci]

Manual promotion looks fine.

[mcara]

I wonder why are 'PSCALE' numpy.float32 at all?

[mcara]

I wonder whether it would be better to modify

stwcs/stwcs/distortion/mutil.py

Lines 200 to 213 in 5a9212e

	refpix = {}
	refpix['XREF'] = ftab[1].data.field('XREF')[row]
	refpix['YREF'] = ftab[1].data.field('YREF')[row]
	refpix['XSIZE'] = ftab[1].data.field('XSIZE')[row]
	refpix['YSIZE'] = ftab[1].data.field('YSIZE')[row]
	refpix['PSCALE'] = round(ftab[1].data.field('SCALE')[row], 8)
	refpix['V2REF'] = v2ref
	refpix['V3REF'] = v3ref
	refpix['THETA'] = theta
	refpix['XDELTA'] = 0.0
	refpix['YDELTA'] = 0.0
	refpix['DEFAULT_SCALE'] = True
	refpix['centered'] = False
	refpix['skew_coeffs'] = skew_coeffs

to simply store Python float instead of whatever type is in the table.

[mcara]

I feel it would be better to just convert tab values to floats right away (currently they can be both np.float32 and Python float (0.0) which is a little inconsistent) then to promote to float64 at the end stage but it is not wrong.

[s-goldman]

I feel it would be better to just convert tab values to floats right away (currently they can be both np.float32 and Python float (0.0) which is a little inconsistent) then to promote to float64 at the end stage but it is not wrong.

Thanks for reviewing this. I would prefer to do it the right way. I couldn't find where they were instantiated, so now that you've pointed it out, I will set the type at the beginning of the process.

[mcara]

LGTM. Thank you!