backlog - better document new data access with recent changes

[anthony-chaudhary]

context that some of the docs are still geared towards file based export

1. need to more clearly defined the two paths (JSON file, streaming)
2. the two major types (Tasks, Datasets)
3. instance/annotation access when streaming and not using to_pytorch() type methods
4. better promote streaming as the default, and the JSON file as secondary/special case access