[Feature][AutoMM] Add Test-Time-Augmentations (TTA)

Test-time augmentations are very useful for improving the vision model performance.

We can offer test-time augmentation to AutoMM via another `tta` flag in the `.predict()` API:

```python
predictor = AutoMMPredictor(...)
predictions = predictor.predict(test_data, tta=tta_config)
```

@zhiqiangdon @bryanyzhu @Linuxdex @lzcemma

Start by locating AutoMMPredictor.predict and reading the existing AutoMM vision prediction flow. The issue names no files or tests, so first establish where prediction options are handled and how the TTA configuration should be represented. Done means predict(test_data, tta=tta_config) is supported and the vision-model behavior is covered by appropriate validation.

Written by the indexing model from the issue text.