Gemma 4 models on Bedrock Mantle do not forward reasoning effort

**What problem would this solve?**

PR #10297 (open) routes `google.gemma-4-31b`, `google.gemma-4-26b-a4b`, and `google.gemma-4-e2b` to the Bedrock Mantle Responses endpoint. All three variants support the `reasoning` parameter on that endpoint ([AWS blog](https://aws.amazon.com/blogs/machine-learning/introducing-gemma-4-models-on-amazon-bedrock/)). Setting `GOOSE_THINKING_EFFORT` has no effect on any of them today: the gate that emits the reasoning block checks whether the model is an OpenAI Responses model, which requires an `openai.` prefix. Gemma model IDs carry a `google.` prefix, so the check always returns false and the reasoning parameter is never sent.

**What would a good outcome look like?**

When `GOOSE_THINKING_EFFORT` is set and the active model is `google.gemma-4-31b`, `google.gemma-4-26b-a4b`, or `google.gemma-4-e2b`, the provider forwards the effort value (`low`, `medium`, or `high`) as the `reasoning` parameter in the Mantle Responses API request. The AWS blog notes that `high` is recommended for E2B specifically.

**Possible approaches**

Extend or complement the existing OpenAI Responses model check in `crates/goose/src/providers/bedrock.rs` to also match `google.gemma-4-*` model IDs, so the reasoning parameter is included when those models are routed through Mantle.

**Additional context**

- PR #10297: introduces Mantle routing for Gemma 4 (prerequisite, currently open)
- Issue #10004: the reachability bug PR #10297 fixes
- [AWS blog: Gemma 4 on Amazon Bedrock](https://aws.amazon.com/blogs/machine-learning/introducing-gemma-4-models-on-amazon-bedrock/) — documents the `reasoning` parameter and per-variant recommendations

- [x] I have verified this does not duplicate an existing feature request

Do not begin implementation until the issue reaches **Ready** on the [Goose Issues board](https://github.com/orgs/aaif-goose/projects/1).