Support negative index in SimpleVideoDecoder

[Dan-Flores]

This PR adds the missing functionality described in #150, specifically adding support for negative indexing in get_frame_at() and get_frames_at().

In both functions, a negative index n is converted to len(decoder) - n. The tests are updated to reflect this support as well.

The other item in that issue, allowing "upper bound in slices to be greated than len(decoder)", is already supported by getitem_slice. Specifically, this section of _video_decoder.py handles negative and out-of-range indices:

    def _getitem_slice(self, key: slice) -> Tensor:
        assert isinstance(key, slice)

        start, stop, step = key.indices(len(self))

I added a new test case to test_getitem_slice to make this more explicit.

[Dan-Flores]

This error is from SingleStreamDecoder.cpp. Should we also be checking valid indices at the Python level, in _video_decoder.py?

[scotts]

Good observation - at the Python level, we currently don't check the input when we get a list of indices (get_frames_at()) or a list of timestamps (get_frames_played_at()). The weak rationale was that we'd have to walk the list to do the check, and that seemed wasteful. Now that we're walking the list in get_frames_at() to deal with negative indices, we might as well do error checking at the same time.

The question, then, is what to do in get_frames_played_at()? For completeness, I think it does make sense to do error checking at the Python level there, but let's do that in a separate PR. We can also do some refactoring there, and make _getitem_slice() call VideoDecoder.get_frames_in_range() in order to get that error checking at the Python level as well.

[Dan-Flores]

It is not necessary to get the expected index here, but I added it to match the other tests. Alternatively, the test can simply match the text: "Invalid frame index".

[scotts]

It's fine to do, but I usually default to just matching the text that doesn't change.

[scotts]

Let's do the error checking in get_frames_at() now too, but approved to unblock.

[NicolasHug]

Thanks @Dan-Flores , LGTM!

I forgot to write it down in #150, but eventually it'd be good to accept the same semantics for get_frames_in_range(). We we can merge this PR as-is though.