ENH: Add MZ3 format

[effigies]

Currently implements reading the header and "scalar" array. Haven't tested if we need to switch ordering to handle multiple scalar arrays. Does not yet implement reading vertices, faces or rgba.

Two significant wrinkles to this format:

1. Gzipped and uncompressed versions have the same extension.
2. The data array size must be calculated from the file size. For the compressed case, I first try to use the member-size mod 2**32 field in the gzip footer to detect the size. If we hit 20GB we bet that it's a weird file and just seek to the end of the gzip stream.

Loading face/header data will be a good test for BIAP 0009 (#1056), but don't want to clog up that PR at the moment.

[pep8speaks]

Hello @effigies, Thank you for updating!

• In the file nibabel/surfice.py:

Line 175:1: W293 blank line contains whitespace

To test for issues locally, pip install flake8 and then run flake8 nibabel.

Comment last updated at 2021-11-19 13:45:48 UTC

[codecov]

Codecov Report

Merging #1066 (fdd94d1) into master (4703f4d) will decrease coverage by 0.77%.
The diff coverage is 28.76%.

@@            Coverage Diff             @@
##           master    #1066      +/-   ##
==========================================
- Coverage   92.30%   91.53%   -0.78%     
==========================================
  Files         100      101       +1     
  Lines       12223    12363     +140     
  Branches     2194     2221      +27     
==========================================
+ Hits        11282    11316      +34     
- Misses        617      721     +104     
- Partials      324      326       +2     

Impacted Files	Coverage Δ
nibabel/surfice.py	25.92% <25.92%> (ø)
nibabel/openers.py	92.56% <42.85%> (-3.17%)	⬇️
nibabel/arrayproxy.py	100.00% <100.00%> (ø)
nibabel/streamlines/array_sequence.py	97.95% <0.00%> (-0.69%)	⬇️

---

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...fdd94d1. Read the comment docs.

[neurolabusc]

@effigies thanks for supporting this. The mz3 specification is here.

• It is true that a single extension is used for both compressed and uncompressed data. You can discriminate these by looking at the first two bytes of the file. Specifically, all GZip files begin with hex = 0x1F8B, all uncompressed MZ3 files begin with hex = 0x4D5A
• You do not need to know the file size to compute the array size. The header NFACE and NVERT explicitly describe the number of indexed triangles (faces) and spatial points (vertices).

Happy to work directly with you to tidy this up.