Feat/optimize model gateway

[mnvsk97]

1. Avoid creating a model instance for every request and instead cache by model name, config, and other metadata.
2. Add simple local dict based cache for embedding, llm, reranker, and audio models.
3. Always check in cache before creating an instance of a model to support 1.
4. Add cachetools library to support simple caching mechanisms and also for complex cases in the future.
5. Add documentation for each method in the file