ENH: Speed up tractogram.apply_affine

[MarcCote]

This PR adds the inplace option to affines.apply_affine(...) which Tractogram.apply_affine can leverage for some speed and memory gains.

The speedup is actually two-fold:

1. Allowing affines.apply_affine to perform the transformation in-place reduces memory peak usage (esp. when dealing with 10 million streamlines).
2. If the ArraySequence representing the tractogram's streamlines is not a sliced view, the affine transformation can be applied directly on the underlying numpy array that contains the data. Doing that avoids having to transform each streamline one by one as it is currently done.

Also, this PR partially addresses issue #943 (see my comment over there for more details).

[codecov]

Codecov Report

Merging #1000 (8c20100) into master (4703f4d) will decrease coverage by 0.07%.
The diff coverage is 100.00%.

❗ Current head 8c20100 differs from pull request most recent head cb037d0. Consider uploading reports for the commit cb037d0 to get more accurate results

@@            Coverage Diff             @@
##           master    #1000      +/-   ##
==========================================
- Coverage   92.30%   92.23%   -0.08%     
==========================================
  Files         100      100              
  Lines       12223    12176      -47     
  Branches     2194     2131      -63     
==========================================
- Hits        11282    11230      -52     
- Misses        617      618       +1     
- Partials      324      328       +4     

Impacted Files	Coverage Δ
nibabel/affines.py	100.00% <100.00%> (ø)
nibabel/streamlines/array_sequence.py	99.30% <100.00%> (+0.66%)	⬆️
nibabel/streamlines/tractogram.py	99.68% <100.00%> (+<0.01%)	⬆️
nibabel/filebasedimages.py	87.23% <0.00%> (-2.13%)	⬇️
nibabel/loadsave.py	92.59% <0.00%> (-1.26%)	⬇️
nibabel/nicom/csareader.py	87.40% <0.00%> (-0.75%)	⬇️
nibabel/volumeutils.py	93.46% <0.00%> (-0.71%)	⬇️
nibabel/openers.py	95.09% <0.00%> (-0.63%)	⬇️
nibabel/nifti1.py	92.10% <0.00%> (-0.47%)	⬇️
nibabel/dft.py	80.98% <0.00%> (-0.31%)	⬇️
... and 12 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4703f4d...cb037d0. Read the comment docs.

[MarcCote]

@effigies @arokem I forgot about this PR until now. Some TRK's header issue was brought up in this discussion: https://neurostars.org/t/trk-files-in-dipy-header-file-incorrect/20452/6. I'm not sure if the issue is related to Dipy/NiBabel or some other library.

[arokem]

This looks good to me!

[effigies]

Sorry for the extremely slow review. This LGTM.

[MarcCote]

@effigies no worries. Happy it's in before 4.0.0 :).