fix: apply custom provider context limits consistently

[d3f4w2]

Summary

• resolve model configs against the active provider registry before doctor requests
• apply the same provider metadata fallback when a session has no persisted model config
• preserve existing behavior for providers that are not present in the registry

Custom/declarative providers already expose per-model context_limit values through ProviderEntry::normalize_model_config, but the doctor and session fallback paths constructed a bare ModelConfig and skipped that normalization. As a result, large-context custom models could silently fall back to 128k.

Testing

• cargo test -p goose providers::init::tests --lib -- --nocapture (8 passed)
• cargo clippy -p goose --all-targets -- -D warnings
• cargo fmt --all -- --check
• git diff --check

All Rust checks ran in the repository toolchain version (Rust 1.96.1).

Related Issues

Fixes #10805

Screenshots/Demos (for UX changes)

Not applicable.

[alexhancock]

Thanks for the submission! It's a valid problem and we have a few other aspects to resolve as well

We're tracking things here for a unified fix #10966