Vocentering gpu offload

[neon60]

Improve performance of vo search

• Use full mask
• Use modified _calculate_metrics

I will validate these again:

• The following CPU implementation has no significant effect on performance, that's why I did not port to GPU: https://github.com/DiamondLightSource/httomolibgpu/blob/vocentering_gpu_offload/httomolibgpu/recon/rotation.py#L191-L202
• The downsample kernel also did not give significance improvement.

Please ignore these files, I will delete them:

• docs/source/metrics_calculate.py
• docs/source/rotation_new.py

[dkazanc]

thanks @neon60 I'm going through it now. We added real data tests into the library recently so I'll add that. Also will clean some comments as I go.

[namannimmo10]

You must add a label to run the Zenodo tests - with 6ae3cb1 the big data tests didn't run. The tests will run only after you add the label and make a commit.

[namannimmo10]

Currently the CI fails because the performance test was collected

FAILED zenodo-tests/test_recon/test_rotation.py::test_center_vo_i12_dataset1_performance