make the large language model abuse throttling parameters configurable

[williamstein]

See this code

https://github.com/sagemathinc/cocalc/blob/master/src/packages/server/llm/abuse.ts

Instead of these being hardcoded, make them configurable in admin settings.

This would be especially useful for on prem users that don't need limits. E.g., https://github.com/sagemathinc/cocalc-docker/issues/218