fix: check model status before inferencing

[vansangpfiev]

Describe Your Changes

This pull request introduces several changes to the InferenceService class and related methods to improve model handling and logging. The most important changes include adding logic to handle model loading, updating logging levels, and modifying the storage of saved models.

Improvements to model handling and storage:

• engine/services/inference_service.cc: Added logic to check if a model is loaded and start loading it if not. This includes retrieving the model status and initiating the model loading process if necessary.
• engine/services/inference_service.cc: Removed redundant code for retrieving the model_id inside a nested block and consolidated it at the beginning of the function.
• engine/services/inference_service.cc: Added logic to save models in the saved_models_ map when they are loaded, ensuring they can be reused later.
• engine/services/inference_service.h: Introduced a new SavedModel type and an unordered_map to store saved models, facilitating efficient model retrieval.

Logging improvements:

• engine/services/inference_service.cc: Changed the logging level from CTL_INF to CTL_DBG for the JSON body inference message to reduce log verbosity.

Fixes Issues

• idea: cortex should auto-reload model when start chat completion #1863

Self Checklist

• Added relevant comments, esp in complex areas
• Updated docs (for bug fixes / features)
• Created issues for follow-up changes or refactoring needed