Fix inconsistent model training approach in trian_and_predict function

[sohampirale]

Problem

The trian_and_predict function in app/services/gaze_tracker.py had inconsistent logic for determining which models should use GridSearchCV with hyperparameters versus simple model fitting:

• "Elastic Net" and "Support Vector Regressor" had hyperparameters defined in config.py but were using simple model fitting
• "Random Forest Regressor" had no hyperparameters defined in config.py but was attempting to use GridSearchCV

---

Solution

Update the conditional logic in trian_and_predict to properly align model usage with hyperparameter availability:

• Models without hyperparameters (Linear Regression, Random Forest Regressor) now use simple fitting
• Models with hyperparameters (Elastic Net, SVR, Ridge, Lasso, Bayesian Ridge, SGD) now use GridSearchCV

---

Impact

This change ensures that:

1. Models with hyperparameter configurations properly utilize GridSearchCV for optimization
2. Models without hyperparameter configurations use efficient simple fitting