[CI] Build and package Treelite with OpenMP

[hcho3]

• Rewrite ParallelFor using OpenMP constructs.
• Link Treelite libs with OpenMP runtime lib.
• On MacOS, bundle libomp (OpenMP runtime) with Treelite. This is to ensure that Treelite does not randomly crash. [CI] Build Python wheels for MacOS (x86_64 and arm64) xgboost#7621 for full explanation.
• Build PyPI wheel targeting Apple Silicon. Closes Apple Silicon support #350
• On Linux, run auditwheel repair command to vendor (bundle) libgomp.so inside the Python wheel. This is required by the Python packaging standard.

[codecov]

Codecov Report

Merging #353 (5bb230b) into mainline (ab699e7) will decrease coverage by 0.36%.
The diff coverage is 62.66%.

@@              Coverage Diff               @@
##             mainline     #353      +/-   ##
==============================================
- Coverage       85.01%   84.65%   -0.37%     
  Complexity         44       44              
==============================================
  Files             108      109       +1     
  Lines            8461     8457       -4     
  Branches          480      474       -6     
==============================================
- Hits             7193     7159      -34     
- Misses           1244     1274      +30     
  Partials           24       24              

Impacted Files	Coverage Δ
src/threading_utils/parallel_for.h	55.55% <53.44%> (-44.45%)	⬇️
tests/cpp/test_threading_utils.cc	94.11% <90.00%> (-2.23%)	⬇️
src/annotator.cc	91.17% <100.00%> (+0.10%)	⬆️
include/treelite/omp.h	50.00% <0.00%> (ø)

---

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 ab699e7...5bb230b. Read the comment docs.