improve read_object speed when passing idx

[lvarriano]

See #29 for details of issue and fix. "Fixes" this issue by sacrificing a small amount memory for additional speed. This can improve read_object speeds when using the idx parameter by more than x50 depending on how many events are selected from a file.

There is now a x2 speed penalty if the user passes in idx that corresponds to all events compared to just reading everything without any idx. In the future, read_object could check if the user did this. For now, however, it is up the user to check.

[codecov]

Codecov Report

Attention: 3 lines in your changes are missing coverage. Please review.

Files	Coverage Δ
src/lgdo/lh5_store.py	82.70% <85.00%> (+0.14%)	⬆️

📢 Thoughts on this report? Let us know!

[gipert]

We should still allow directly reading with slice objects. How would the user do that?

[gipert]

Ok nevermind, slices are used when setting start_row/n_rows.

[lvarriano]

Thanks for the idea, Luigi! I changed it so that if idx can be converted to a slice instead (all indices are increasing by 1), then it will do that. This restores some of the speed if the user happens to pass in idx that corresponds to all events (now ~30% slowdown instead of x2). And this improved the speed of the original implementation.

[jasondet]

looks very good to me. I'm going to merge.