Change the types of PySliceIndices and `PySlice::indices

[cmpute]

This fixes #3756. I decided not to change the type of length to usize, to leave the user with the option to skip the check, because isize::MAX is already very large as a length of sequences.

[codspeed-hq]

CodSpeed Performance Report

Merging #3761 will not alter performance

_{Comparing cmpute:slice_index (2679a28) with main (c1f11fb)}

Summary

✅ 70 untouched benchmarks

[davidhewitt]

Thanks! Overall I think this is a straightforward thing for users to update for, and makes things a bit tidier, so overall I'm warm to this change.

My only hesitation as I mentioned before is migrating from 0.20 to 0.21 is already set to be a big job for users so I'm a bit wary of adding more breaking changes right at this minute, especially as this doesn't seem strictly necessary.

I'd be happy to hear other maintainer's views here; otherwise if you can forgive me @cmpute, I'd prefer to take a step of caution and save this for 0.22.

[cmpute]

I'm open for the merge timepoint :) Waiting for 0.22 sounds good to me!

[cmpute]

I have rebased the code, not sure why the codecov check failed tho.

[davidhewitt]

Thanks, sorry to take a while to get to this, let's merge!

The uncovered path looks like it's the old GIL Ref API, we'll live with that for the moment. (Going to be nice in 0.23 when coverage shoots back up!)

[jessekrubin]

I am using slices, and this is causing me some trouble in upgrading to 0.22. do I HAVE to enable gil-refs just need those slice indices?

[davidhewitt]

No, you need to use the new PySliceMethods trait.