Add ProsAudit benchmark

[nhamilakis]

• [x] Create a new Dataset
• [x] Create a new Prosody Task (based on sBlimp (Syntactic)
• [x] Create a new benchmark running task with specific dataset
• [x] Create submission for prosody task
  • [x] Create form of submission
  • [x] Create validation for submission
  • [x] Create score dir
  • [x] Add leaderboard schema
• [x] Add leaderboard file generation

[nhamilakis]

With this a2faf4db936c6f4a3b62adb0c4065dfac06b9977 commit, i added the basic integration layout for the new benchmark.

• added Dataset class
• added Task class
• added Submission class
• added Benchmark class

When task is more clear, add :

• eval implementation
• dataset index
• submission validation
• score & leaderboard implementation
• export the benchmark

[nhamilakis]

With this edf89260609d94a81f24866d4a26727af5672696 commit added the continuation of a2faf4db936c6f4a3b62adb0c4065dfac06b9977

• added finished task copied from syntactic.
• added submissions
• added submission validation
• added score_dir object

TODO:

• [x] Checkout bug with prosody_by_pair, columns in gold have been changed from syntactic creates issues here

https://github.com/zerospeech/benchmarks/blob/edf89260609d94a81f24866d4a26727af5672696/zerospeech/benchmarks/tasks/lm/prosody.py#L34-L45

- [ ] Leaderboard entry (once scores have been evaluated)

[nhamilakis]

3f51bc122f62b909e9143109e99d5204969fdd4b fixes mismatch on prosody scoring by_pair.

Syntactic benchmark had transcription and by_pair scoring was done using transcriptions as id. For prosody benchmark transcription column is null, so by_pair scoring needs to contain id column to be identifiable.

[nhamilakis]

d58a500f014915a9884387870ac46f45b9d71bef added leaderboard schema

[nhamilakis]

d63e0585c236e92581ecf6b49ce3e9ae4bc79576 added leadereboard generation