[Lgbm] support multi classification

[ewan0x79]

Description

#3223

This PR introduces an enhancement to the LightGBM inference functionality, specifically targeting multi-classification tasks. It modifies the inferenceMat method to dynamically allocate output buffer size based on the number of classes in the model, thus enabling support for multi-class predictions.

Key Changes:

• The method now retrieves the number of classes in the model using LGBM_BoosterGetNumClasses and allocates the output buffer (outBuffer) accordingly. This ensures that the buffer size is sufficient for storing prediction results for all classes.
• This change is backward compatible as it extends the existing functionality without altering the method signature or expected input/output types. It enhances the method's capability to handle multi-classification tasks, which was previously not explicitly supported.