Lock pipeline dependencies to sensible semversions.

[emirkmo]

The photometry pipeline needs to be cleaned from the API and other parts, and should be locked to sensible but not too old semversion for the dependencies. For example, currently it's not possible to pip install astropy 4.1 on newer python version because of incompatibility with Jinja2 version in its build dependencies. If we don't have a reason for staying on 4.1, we should test moving up, at least to 4.3, and later to 5.0. Astropy versions are technically LTS supported for 2 years.

However, we should also lock the dependencies, make only 1 version of python a required CI check (because this will be a pipeline/app that is locked and using that version). We can over time bump that python version (we don't have to lock the python version though). We can continue testing against py3.9 and py3.10, in preparation for moving to them, but don't need to require it to be working.

• astropy v4.3x? 5.0x? Upgrade to 4.2x-4.3x would mean that older version ECSV tables cannot be read. 5.0x has this solved, perhaps this can be back ported or we can upgrade to 5.0, not sure if breaking anything or changing results, (especially for phot utils). Needs testing for 5.0 due to modeling changes.
• numpy (latest?): 1.22+ some issues with building on certain Macs, maybe need to keep to >1.19.1 <1.22
• matplotlib (latest?), staying at 3.3 for now but need to move.
• scikit-image (newer supported version?) => 0.14.2
• scipy (latest?): 1.6.0+, latest seems fine.
• Photutils (V1.3 ideally, but start with V1.2):
  Looks like Photutils can be moved to V1.2 with zero changes, or to V1.3 by making sure to catch the possible ValueErrors for misconfigured initial guesses. This is fine on astropy 4.0x+ but needs a bump to spicy 1.6.0, scikit-learn 0.19, scikit-image 0.14.2, gwcs 0.12.
• Bottleneck (latest should be fine).

[emirkmo]

I ran a bunch of tests on test_ci branch, and this Jinja2 issue is very annoying to workaround.

I can actually get it working on ubuntu, but astropy fails to build on Windows, and sci-kit image fails on Mac OS for py>3.9.

[emirkmo]

Astropy can be Brought to 5.1 with the pending change to g2d fitting.

Numpy, matplotlib, scipy, scikit learn be brought to latest according to the test, pending the results from the "master" log.