[Feature] Cache for LLMs

[deep-learning-dynamo]

As a java developer I want to be able to cache responses from LLM:

• to speedup frequent prompts
• to save noney

Requirements:

• easy way to plug in custom cache providers
• simple cache: match the prompt with simple heuristics: ignoring case / ignoring whitespaces / etc
• semantic cache: by semantic similarity with configurable threshold