Support more NWB training data types

[keyaloding]

As written in #86, some NWB data types are not supported yet in sleap-io, including TrainingFrame, TrainingFrames, SourceVideo, and PoseTraining. #24 shows the data model equivalences between SLEAP and NWB, and I am expanding support for NWB data by allowing sleap-io to save and load it. I have written a function convert_nwb in nwb.py that initializes new SLEAP objects when an NWB object is used as an argument.
https://github.com/keyaloding/io_fork/blob/70a34ede294d3c783d1758c55c1dd9752e29f1e7/sleap_io/io/nwb.py#L37-L72

I am planning to write a new function that calls the read_nwb in sleap_io/io/nwb.py and returns the attributes of the Labels object that is returned. It will return either Labels.labeled_frames, Labels.videos, or https://github.com/talmolab/sleap-io/blob/5a301106effd8cbcc6f23c9385e3c274b6b9262f/sleap_io/io/nwb.py#L40-L48 These attributes will correspond to theTrainingFrame, TrainingFrames, SourceVideos, and PoseTraining` data types. I will add new tests that run inference and training on .nwb data