perf: speed up rotation by about 20%

[nicrie]

Move some computations in front of the loop and use broadcasting instead of matrix multiplication. I ran some tests that showed an improved performance of about 20 % on average. I also tried numba but that did not bring any significant gains.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 4 lines in your changes missing coverage. Please review.

Please upload report for BASE (develop@7cba749). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
xeofs/linalg/_numpy/_rotation.py	0.00%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop     #228   +/-   ##
==========================================
  Coverage           ?   24.65%           
==========================================
  Files              ?       58           
  Lines              ?     3970           
  Branches           ?        0           
==========================================
  Hits               ?      979           
  Misses             ?     2991           
  Partials           ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[nicrie]

@slevang just in case you have some smart ideas how to make this rotation process more efficient.

[slevang]

nope, no ideas :) looks good to me. In my (limited) experience, rotation is pretty fast, and good enough, if you accept a lower convergence tolerance.