MNT: Refactor Docker build process

[effigies]

Changes proposed in this pull request

• Break Docker build into stages, with each significant (and often slow) download getting its own stage
• Switch conda environment to being micromamba-managed
• Create env.yml with the pinned conda environment, keeping this information outside of the Dockerfile
  • Due to FSL relying on portions of the scientific python stack, we continue to install that from conda
• Use pip-compile to pin our dependency graph and install that after the conda environment

This means that env.yml and requirements.txt need to be updated in lock-step if the dependency affects env.yml. We can probably script this.

Documentation that should be reviewed

[effigies]

@mgxd @oesteban Here's a working proposal. ds005 is expected to fail due to ICA-AROMA and FSL 6.0.6.2.

[effigies]

Test failures: nipreps/niworkflows#795.

[mgxd]

Looks like something is wrong with ds210's SDC

Traceback:
	Traceback (most recent call last):
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/nipype/interfaces/base/core.py", line 398, in run
	    runtime = self._run_interface(runtime)
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/niworkflows/interfaces/nibabel.py", line 467, in _run_interface
	    self._results["out_file"] = _advanced_clip(
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/niworkflows/interfaces/nibabel.py", line 705, in _advanced_clip
	    a_min = np.percentile(
	  File "<__array_function__ internals>", line 200, in percentile
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/lib/function_base.py", line 4205, in percentile
	    return _quantile_unchecked(
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/lib/function_base.py", line 4473, in _quantile_unchecked
	    return _ureduce(a,
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/lib/function_base.py", line 3752, in _ureduce
	    r = func(a, **kwargs)
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/lib/function_base.py", line 4639, in _quantile_ureduce_func
	    result = _quantile(arr,
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/lib/function_base.py", line 4745, in _quantile
	    take(arr, indices=-1, axis=DATA_AXIS)
	  File "<__array_function__ internals>", line 200, in take
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/core/fromnumeric.py", line 190, in take
	    return _wrapfunc(a, 'take', indices, axis=axis, out=out, mode=mode)
	  File "/opt/conda/envs/fmriprep/lib/python3.10/site-packages/numpy/core/fromnumeric.py", line 57, in _wrapfunc
	    return bound(*args, **kwds)
	IndexError: cannot do a non-empty take from an empty axes.

[effigies]

I guess we can try bumping caches...

[effigies]

@mgxd I canceled ds005 just because it takes so long. Otherwise this is passing. Note the codecov removal is due to https://community.codecov.com/t/codecov-yanked-from-pypi-all-versions/4259, and has no additional impact since we haven't been using that package.

[mgxd]

A few suggestions, but this is really nice!

One consideration (not needed for this PR) would be to create these build stages into their own dedicated repos in nipreps-containers, so other tools could easily grab what they need with a simple COPY --from=<nipreps/tool> ...

Closes #2441

[mgxd]

Is this handling the extras too? It should account for [telemetry,test]

[effigies]

I'm okay adding telemetry, but the function of the tool does not depend on the versions of test dependencies, so I'd rather not set hard pins.

[mgxd]

should fix it, my bad!