feat: Add dedicated time-series cross-validation pipeline

- Add `grid_search_cross_validate_ts.py`, a new pipeline module tailored for time-series models from the `aeon` library. This separates TS logic from the standard tabular pipeline.

- Implement extensive monkey-patching for `aeon` classifiers (`_patch_aeon_models`) to enhance stability and fix common issues encountered during hyperparameter search:
    - Fix `AttributeError: '_metrics' missing` in deep learning models by ensuring state is correctly initialized during `fit`.
    - Add automatic data padding to prevent Keras shape-related errors (e.g., "negative output size") with short time series.
    - Stabilize predictions by replacing NaNs from unstable models with a uniform distribution.
    - Resolve parameter conflicts and `IndexError` crashes in `MUSE` and `ResNet`.

- Refactor `model_class_list_ts.py` to dynamically load models from the configuration file using a `TS_MODEL_CLASS_MAP`, replacing the previous hardcoded list.

- The new pipeline forces a "threading" backend for `joblib` and single-threaded execution for GPU/Keras models to prevent pickling errors and multiprocessing hangs.

Author: SamoraHunter