fix(provider): require optional provider model id (#620)

[Daedalus-Icarus]

Summary

Fixes #620.

Parameterize the optional OpenAI-compatible provider path with an explicit model id instead of letting it fall back to the generic "default" placeholder.

Changes

• add [optional_ai_provider].model to config and require it when the optional provider path is enabled
• thread the configured model through LlmClient::from_config() into the OpenAI-compatible backend
• serialize that configured model in outbound generic OpenAI-compatible chat requests
• update provider/config/runtime-boundary test fixtures and docs/config templates

Real Behavior Proof

• I have built and run the affected code locally (or noted why I could not).
• I have verified the change end-to-end on Jetson hardware.
• I have NOT verified on Jetson hardware, and I explain the equivalent verification path or validation gap below.

Tested profile / hardware (check all that apply):

• jetson
• raspberry_pi
• portable_sbc
• laptop
• mac
• CI-only / docs-only
• Not run locally

What I ran

cargo fmt --all -- --check
cargo test -p genie-common validate_llm_provider_rejects_enabled_optional_provider_without_model
cargo test -p genie-common optional_ai_provider_can_use_oauth_bearer_env
cargo test -p genie-core --test provider_config_test from_config_selects_optional_openai_compatible_provider
cargo test -p genie-core generic_request_profile_uses_supplied_model_and_keeps_full_body
cargo test -p genie-core local_openai_compatible_provider_can_be_ready

What I observed

• optional-provider config now fails loudly when [optional_ai_provider].model is empty
• the generic OpenAI-compatible request-profile test now verifies that the supplied model id is emitted instead of "default"
• the optional provider selection and readiness tests still pass with an explicit configured model id
• I did not verify this on Jetson because the change is provider-config/client plumbing and documentation rather than a hardware-facing runtime path

Test plan

• Configure [optional_ai_provider] with a local mock OpenAI-compatible endpoint and a real model value.
• Confirm startup rejects an empty model.
• Capture one outbound /v1/chat/completions request and verify the configured model id is present in the JSON body.

Notes for reviewers

This is a narrow follow-up to the merged provider-selection work from #595: provider selection was config-driven, but provider model selection was still hardcoded in the generic request profile.

[matedev01]

Reviewed — correct fix for #620 (model="default" hardcoding), clean plumbing. Please finalize: unmark draft, tick the proof section, and rebase (BEHIND). Ready to merge once green.