Akd service

[leothomas]

Apologies for the reformatting related line changes - I didn't see a pre-commit in the repo, but happy to re-run with pre-commit to reduce the number of changed files.

Code changes:

• Rename corutine --> coroutine
• Update NLI label extraction to use regex to search for neutral , entailment , contradiction within square brackets in the last position. (some models insert a period or other symbol after the label within the brackets, or sometime double brackets, which can cause the current label extraction to fail)
• Add configuration for requesting log_probs for LiteLLM and OpenAI request formats
• Update token extraction logic to handle BedRock (LiteLLM) and Ollama style token structure (poorly standardized across model providers, unfortunately)
• Add threading for ContextRetrieval for improved runtime
• Add helper function to produce graph execution as json

[leothomas]

(Note: I opened this PR against main rather than akd_service as I have rebased my fork off of main to keep it updated - this PR is huge against akd_service, but if someone can update akd_service from main, I can update this PR to merge into akd_service, if preferred