Add trim_to_alignments() method

[desh2608]

Based on discussion in #923

[lubacien]

All tests are passing on my side (although I have to use torch.set_num_threads(1) to pass the test test_compute_and_store_features_lazy(nj) with nj=2, this is not related).
I also tested it from command line on my data and it works as expected.

[desh2608]

A catch here is that if there are alignment items which are continuous (i.e. prev_end == cur_start) then they will necessarily get merged. So in order to trim such that we get #cuts = num. of alignment items, a negative value of max_pause will need to be passed.

[desh2608]

All tests are passing on my side (although I have to use torch.set_num_threads(1) to pass the test test_compute_and_store_features_lazy(nj) with nj=2, this is not related). I also tested it from command line on my data and it works as expected.

Thanks for testing this. I will verify it on my end and then ask @pzelasko to merge if he doesn't see any immediate issues.

[desh2608]

I think I have convinced myself that this works as expected.

@pzelasko could you please check when you have some time?

[pzelasko]

Thanks! Looks good to me, just left one comment.

[desh2608]

Thanks! Looks good to me, just left one comment.

Done.