feat: Add Cached Cost in calculation

[Zerka30]

Hello 👋 ,

I am writing to propose a new feature for OpenWeb UI Monitor that would allow users to take into account the caching costs of requests made to various large language model (LLM) APIs, such as OpenAI and Anthropic.

As many users rely on these APIs for their projects, it is essential to consider the associated costs, especially when dealing with long conversations that can quickly become expensive without proper caching. Currently, OpenWeb UI Monitor does not provide an option to automatically structure prompts to use cache_control, which can lead to higher costs for maintainer.

[VariantConst]

Caching isn't guaranteed to be hit, so even if we treat overlapping tokens as cached, the actual cost might still vary significantly. I don't think this is the correct way to deal with caching.

Certain model providers like OpenAI and DeepSeek include a cached tokens field in their JSON responses, but it’d take some manual effort to dig into the specific response formats for each provider. I’ve been pretty busy recently, so it’d be a huge help if you could contribute by identifying the exact fields related to cached tokens of each provider (including but not limited to OpenAI, Anthropic, Gemini, DeepSeek, Grok, ...). Otherwise, it might take a while before I can get to this feature myself.