Update groq and Gemini default models

groq are [deprecating `llama-3.1-8b-instant`](https://console.groq.com/docs/deprecations#deprecation-history) in August and recommend migrating to `openai/gpt-oss-20b` I did a quick scan to see where chatlas and ellmer diverge, and the only one which I saw was that default Google Gemini model was recently updated to `gemini-3.5-flash` in [ellmer](https://github.com/tidyverse/ellmer/pull/1026), but is currently 2.5 in chatlas.