Model chat: allow multiple final ratings

[EricMichaelSmith]

Patch description
In the model chat crowdsourcing task, ask for and save multiple 1-to-5 ratings at the end of the HIT. Questions can be specified through mephisto.blueprint.final_rating_question, with multiple questions separated by a pipe character. Multiple ratings will be saved in the format '2|4|3'.

Testing steps
With 1 final question:

python parlai/crowdsourcing/tasks/model_chat/run.py \
mephisto.architect.port=3006 \
mephisto.task.maximum_units_per_worker=1000000 \
mephisto.task.task_name=${TASK_NAME} \
mephisto.blueprint.num_turns=2

With multiple final questions:

python parlai/crowdsourcing/tasks/model_chat/run.py \
mephisto.architect.port=3006 \
mephisto.task.maximum_units_per_worker=1000000 \
mephisto.task.task_name=${TASK_NAME} \
mephisto.blueprint.num_turns=2 \
mephisto.blueprint.final_rating_question=\'Please\ rate\ your\ partner\ on\ a\ scale\ of\ 1-5.\|On\ a\ scale\ of\ 1\ to\ 5,\ what\ is\ your\ favorite\ number?\'

[JackUrb]

This is a cool additional feature! Happy to approve.

From an API design perspective, I wonder about the choice to |-separate your questions and output vs providing both as an array. I suppose the issue here is backwards-compatibility?

[JackUrb]

Could be:

let all_ratings_filled = ratings.every((r) => r !== "")
let rating = ratings.join('|')

[EricMichaelSmith]

Ha, great - thanks, changed

[EricMichaelSmith]

This is a cool additional feature! Happy to approve.

From an API design perspective, I wonder about the choice to |-separate your questions and output vs providing both as an array. I suppose the issue here is backwards-compatibility?

Yeah, exactly - a bunch of projects seem to currently be relying on model chat, making breaking changes costly, and I'm not convinced of how often we'll want to use multiple final questions anyway. I'm definitely open to changing to the array syntax, which would indeed be more ideal, down the line