docs: correct GPT-5 token limit scope to GitHub Models only

[detail-app]

Updated documentation and comments to reflect that the 7K token input
cap applies only to GPT-5 served via GitHub Models, not the entire
GPT-5 family. GPT-5 models served via OpenAI direct or the LLM gateway
use the 100K default token budget.

Without this correction, a developer might incorrectly assume all GPT-5
models have an 8K input limit and over-truncate prompts for models that
support far larger contexts.

Introduced by commit ce24b66 (@WilliamAGH, #52)

---

_{Doc Drift PRs can be configured here.}