set up Guardrails API to check LLM inputs/outputs

[tatianab]

Add a package that calls the Guardrails API (https://developers.google.com/checks/guide/ai-safety/guardrails), which will be used to check LLM inputs and outputs against policies.

[gopherbot]

Change https://go.dev/cl/637977 mentions this issue: internal/llmapp: allow policy checker in overview functions

[gopherbot]

Change https://go.dev/cl/637976 mentions this issue: internal/gcp/checks: add implementation of llm.PolicyChecker

[gopherbot]

Change https://go.dev/cl/637978 mentions this issue: internal/gaby: turn on policy checks for overviews in Gaby

[gopherbot]

Change https://go.dev/cl/637975 mentions this issue: internal/llm: add interface PolicyChecker

[gopherbot]

Change https://go.dev/cl/637979 mentions this issue: internal/{llmapp,gaby}: display policy evaluation results in web UI