Swap neutral to abstain and have LLM judge yes|no|abstain

[sprice]

This feature changes:

• our judgements from neutral to abstain (an MP can vote yes, no, or abstain so we want to match that)
• moves the abstain judgement to the LLM system prompt rather than having our code swap a yes or no to neutral/abstain by checking if it's a social issue or if one tenet response is different from all others
• for now I've removed the check for one tenet being different than all others (this doesn't make sense to me so if we want this lets unpack it. If all tenets but one are yes it should probably still be a yes?)
• I'm not sure the best way to run mongo migrations so I've created a simple script to run them from a migrations folder. Happy to switch this up if we'd like
• Addresses Change neutral to abstain and update prompt to judge yes, no, or abstain #61

[sprice]

I've run the migration and tested things out and it works as expected.

This one is a non-trivial refactor and would benefit from a careful code review.

[xrendan]

Did you run the migration against the prod database?

I will review this but I will let @mikaalnaik approve/not because he's most familiar with how this all works

[sprice]

No I have not run the migration on prod.

That would happen once this is approved and ready to go. And I should note that this is a fairly rudimentary migration (ie: run a mongo command).

It would be great to get the dev environment up to test this on.

1. Bring dev environment online
2. Sync prod db to dev db
3. Test the deployment of this branch and migration on dev
4. If it looks good, roll it out.

[sprice]

I've added a new script src/scripts/sync-db.ts which can be used to manage MongoDB dumps and restores for both dev and prod databases.

I've also updated the in memory page cache to only be turned on with process.env.NODE_ENV === "production"

[sprice]

src/app/page.tsx still had some logic that swapped the vote in the UI depending on the count of tenet evaluations.

We're now displaying the vote in the UI that the LLM returned. We want clean separation of concerns between how a bill is analyzed and how we display the analysis.