[ENH] Add categorical feature encoding for GFO adapters

[AdityaPandeyCN]

To Resolve #136 This PR adds automatic categorical feature handling for GFO-based optimizers by encoding non-numeric search space dimensions to consecutive integers internally, then decoding them back when evaluating the objective and returning results. Changes:

1. Categorical dimensions (detected via dtype: object, unicode, string, boolean) are encoded to consecutive integers using np.unique
2. Original category mappings are stored in _categorical_mappings for decoding
3. Decoding happens transparently during objective evaluation and when returning best_params_
4. Numeric dimensions are left unchanged
5. Added capability:categorical tag with value "encoded" to _BaseGFOadapter
6. Added test covering mixed numeric/categorical search spaces

[AdityaPandeyCN]

@fkiraly Please have a look, when time permits.