Return finish_reason as an Enum - with wrapped stopping word/sequence/tok

[ShelbyJenkins]

I use stop words/sequences to determine next steps after a response. So if the LLM returns stop word A, we perform action X.

I also do the same with other finish reasons - for example retrying with longer context size if the finish_reason is Length.

Current behavior:

• finish_reason is a string
• stop sequences/words are not returned, so knowing which stop sequence was hit is not possible

Proposed behavior

• finish_reason is an enum. this will allow for easier matching
• the finish_reason stop sequence/word/tok variant includes the string that caused the end of inference