Classify conversations instead of user_messages

[lasryaric]

Description

After doing a first test of classifying 100 user messages, I realized that classifying conversations makes more sense.
A good amount of user messages seems to be just a follow up of a previous user messages, and without the whole list of user messages from the same conversation, it makes no sense to classify them.

We do not classify conversations with more than 30 messages, as they might talk about many different topics.

I also realized that I was missing a few categories.

Risk

Deploy Plan

• deploy front
• run on front DB : DROP TABLE user_message_classifications
• npm run initdb (on front)

[github-actions]

	Warnings
⚠️	Files in front/lib/models/ or connectors/src/lib/models/ have been modified and the PR has the migration-ack label. Don't forget to run the migration from the -edge infrastructure.

Generated by 🚫 dangerJS against 4ae390d

[flvndvd]

Please update the deploy plan to release this PR. We should not rely on Sequelize to alter the table name.

[flvndvd]

Can we extract the await from the if block?

[lasryaric]

Please update the deploy plan to release this PR. We should not rely on Sequelize to alter the table name.

@flvndvd adjusted deploy plan.
I tested and can confirm that the app is not going to crash if we make no call to the user_message_classifications table.