Implement mechanics for intent status response

[matKlju]

AS A Service-Module developer
I WANT server-side and integration logic that maintains a two-state training flag for intents
SO THAT the UI can show a clear green/red indicator and we can block publishing when intents are not trained

Acceptance Criteria

• **Two statuses only:

  • ** TRAINED and NOT_TRAINED.

• **On create

  • ** New intent → status=NOT_TRAINED, needsTraining=true, record lastModifiedAt.

• **On change:

  • ** Any training-relevant edit (examples, entities, responses, rules, synonyms, config that affects model) → status=NOT_TRAINED, needsTraining=true, update lastModifiedAt.

• **On model switch:

  • ** If the active model version changes and the intent is not included in that model for the environment → flip to status=NOT_TRAINED (treat as not trained for that env).

• UI indicator contract:
  - 🟢 Green – Trained ⇢ status=TRAINED and modelVersion[env] == activeModelVersion[env]
  - 🔴 Red – Not trained ⇢ otherwise

[IgorKrupenja]

@matKlju

After fixes are deployed, please additionally test the following. There are quite a few behind the scenes changes:

• Service Module: available intents are correctly shown in Connect intent, there are no regressions.
• Training Module: marking intent for service works, this is reflected in Service module too.
• Training Module: there are no issues and SQL migration failures on deployment.
• Training Module: models can be trained without issues.
• Training Module: editing intents works without issues.
• Training Module: rendering intent list and adding intent is working fine.
• Training Module: On Responses page, adding and editing intents works without issues.
• Training Module: "Teemade ülevaade" page shows model intents without issues.
• Training Module: "Mudelite võrdlus" -> "Activate model" button works well without issues. Models are actually changed.
• Training Module: Deleting, adding and editing rules works without issues.

On model switch logic implemented like this:

• Old Model Intents: Intents that were present in the old model but not in the new model are marked as NOT_TRAINED status, indicating they need to be retrained to be included in future models.
• New Model Intents: Intents that are present in the new model but not in the old model are marked as TRAINED status, but only if they were created before the model training date. This ensures only intents that were actually part of the training process get the trained status.