TEST: Bring back Python 3.8 testing to GHA workflows

[jhlegarreta]

Bring back Python 3.8 testing to GHA workflows.

Removed in commit cf4338d.

Raise the minimum required NumPy version for the Python 3.8
compatibility test to 1.23.0 from the previously required 1.22.0 version
in commit 318bfc6 so that tests can pass. Fixes:

ImportError: numpy.core.multiarray failed to import

raised for example in:
https://github.com/dipy/dipy/actions/runs/6620737573/job/17983666930#step:9:12437

[codecov]

Codecov Report

Merging #2954 (6c3a526) into master (3ad9ec1) will increase coverage by 0.01%.
Report is 109 commits behind head on master.
The diff coverage is 80.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2954      +/-   ##
==========================================
+ Coverage   81.77%   81.79%   +0.01%     
==========================================
  Files         146      146              
  Lines       20401    20401              
  Branches     3238     3238              
==========================================
+ Hits        16682    16686       +4     
+ Misses       2901     2898       -3     
+ Partials      818      817       -1     

Files	Coverage Δ
dipy/data/fetcher.py	40.25% <ø> (ø)
dipy/info.py	100.00% <100.00%> (ø)
dipy/io/stateful_tractogram.py	76.96% <100.00%> (ø)
dipy/reconst/dti.py	92.57% <0.00%> (ø)
dipy/tracking/local_tracking.py	94.44% <96.29%> (ø)
dipy/io/utils.py	88.83% <90.00%> (ø)
dipy/io/streamline.py	84.92% <75.00%> (ø)
dipy/align/imwarp.py	95.60% <37.50%> (ø)

... and 1 file with indirect coverage changes

[skoudoro]

Hi @jhlegarreta,

the minimal py38 is failing due to numpy version. Can you update it ?

[jhlegarreta]

the minimal py38 is failing due to numpy version. Can you update it ?

I've seen it, Serge. Allow me some time, please.

[jhlegarreta]

I've seen it, Serge. Allow me some time, please.

Done.

Codecov failure is unrelated.

[arokem]

LGTM!

[jhlegarreta]

BTW, tried two times (1.22.0 : 1.22.4 : 1.23.0) to raise the min required numpy until tests passed with 1.23.0, this version being greater than the ones required for Python 3.9 and 3.10 (1.22.4)

I'd guess that this numpy incompatibility issue may be related to the circular dependency between trx and DIPY. I agree that it should probably happen with other Python versions as well; not sure it only happens with 3.8. Fixing the circular dependency seems a good step going forward.

[skoudoro]

BTW, tried two times (1.22.0 : 1.22.4 : 1.23.0) to raise the min required numpy until tests passed with 1.23.0, this version being greater than the ones required for Python 3.9 and 3.10 (1.22.4)

I'd guess that this numpy incompatibility issue may be related to the circular dependency between trx and DIPY. I agree that it should probably happen with other Python versions as well; not sure it only happens with 3.8. Fixing the circular dependency seems a good step going forward.

I will tag @frheault, to look at this topic.

For now, thank you @jhlegarreta. merging