Add Assembly AI annotations and gentle files

[complexbrains]

This PR adds friends language annotations run by assembly AI and htk module, the accompanying dataset description file and assembly AI gentle files (that gives word by word onset, offset, duration of each movie segments.)

[bpinsard]

• Unless I am missing something, why should we ship 3 different versions of the same annotation? It seems preferable to ship a single format, the simplest to use and most data efficient ( I would say the tsv that avoid redundant keys compared to json, we can add the confidence to the tsv to avoid losing infos).
• I've randomely seen some character names without capital, maybe that would be worth having a filter to fix that.
• the folder assembly_AI_json contains some tsvs, better not put json in the folder name if that's not all json.
• assembly_AI_json/json_aa/s1 contains a few tsvs.
• The tsvs should have a single sidecar to describe the columns.
• there should be a README in that folder describing how the annotations were produced.
• if we want the files to be better labeled, I would suggest something like task-s01e01a_desc-utterances_events.tsv that better describes the content/format