Add back support for absolute source paths but deprecate it

[astrofrog]

@neutrinoceros - #67 broke compilation for modules with extensions that were specified as absolute paths, which occurred once in astropy (astropy/astropy#15663). The error was:

      writing manifest file 'astropy.egg-info/SOURCES.txt'
      error: Error: setup script specifies an absolute path:
      
          /Users/tom/Code/Astropy/astropy/./astropy/io/fits/hdu/compressed/src/compression.c
      
      setup() arguments must *always* be /-separated paths relative to the
      setup.py directory, *never* absolute paths.

Even though those modules should indeed use relative paths, we should ideally not break compatibility for all packages that were released in the past that use extension-helpers in pyproject.toml without any pinning, so this PR is a patch to restore compatibility but add a warning.

[codecov]

Codecov Report

Attention: 6 lines in your changes are missing coverage. Please review.

Comparison is base (2e72417) 74.68% compared to head (0283551) 62.61%.

❗ Current head 0283551 differs from pull request most recent head 1b2b0f7. Consider uploading reports for the commit 1b2b0f7 to get more accurate results

Files	Patch %	Lines
extension_helpers/_setup_helpers.py	60.00%	6 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main      #70       +/-   ##
===========================================
- Coverage   74.68%   62.61%   -12.07%     
===========================================
  Files           4        4               
  Lines         316      329       +13     
===========================================
- Hits          236      206       -30     
- Misses         80      123       +43     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[neutrinoceros]

Thank you for fixing it, I had no idea my changes could be breaking. Given how important backward compatibility is, I think it'd be useful to have a test for this regression, but I don't know how much effort that'd be to implement and I don't think the lack of a test is reason enough to hold this PR !

[neutrinoceros]

Thinking a bit more about this, it seems like you're making a point that build time dependencies should be pinned more strictly, since every release without pins is bound to break at some point, right ?

[astrofrog]

Yes, see #71

[astrofrog]

The Windows failures are real, I will try and resolve these