Make prediction inserts atomic with auto_commit control and unified service

[extreme4all-ai]

Summary

worker_ml/core.py:31-36 inserts into prediction and prediction_latest tables via two separate repository calls that each commit independently. If the second commit fails, the prediction table is updated but prediction_latest is stale — data inconsistency.

Proposed Change

Two parts:

1. Add auto_commit parameter to database component repository methods

Add an auto_commit: bool = True parameter to repository insert/update methods so callers can control commit timing:

async def insert(self, session, predictions, auto_commit=True):
    ...
    if auto_commit:
        await session.commit()

This enables callers to wrap multiple repo operations in a single transaction.

2. Provide a unified prediction service in the database component

Add a PredictionService (or similar) in components/bot_detector/database/prediction/ that inserts into both prediction and prediction_latest tables atomically. This keeps the transaction control in the component layer instead of the base:

async def insert_predictions(self, session, predictions):
    await self.prediction_repo.insert(session, predictions, auto_commit=False)
    await self.prediction_latest_repo.insert(session, predictions, auto_commit=False)
    await session.commit()

This way bases just call one method and get atomicity for free.