Chat settings never appear: getAvailableConfiguration() ignores getModelSettings(), so reasoning_effort is unreachable

Problem/Motivation

OpenRouterProvider::getModelSettings() adds a reasoning_effort select for reasoning models. But OpenRouterProvider::getAvailableConfiguration() overrides the base method. It returns only dimensions for embeddings and [] for chat, and it never calls getModelSettings(). As a result, the effort setting never appears in provider configuration forms, never gets a default, and is hidden from consumers that read getAvailableConfiguration('chat', $model).

Steps to reproduce

drush ev 'print_r(\Drupal::service("ai.provider")->createInstance("openrouter")->getAvailableConfiguration("chat", "openai/gpt-5-mini"));'

This prints an empty array.

Proposed resolution

1. In getAvailableConfiguration(), return $this->getModelSettings($model_id, $general) for chat, as #3604211: Embeddings operation ignores provider configuration (dimensions parameter not passed) did for embeddings.
2. Replace the hard-coded id prefixes in isReasoningModel() with /models metadata. Each reasoning model has reasoning.supported_efforts and reasoning.default_effort. As of 2026-10-02, 332 of 466 models have them. Build the options from supported_efforts. This covers Claude, Gemini, Grok, DeepSeek and others, not only openai/gpt-5*, o1 and o3.
3. Send reasoning: {effort: …}, which is OpenRouter's documented shape. Top-level reasoning_effort is also accepted. Both are validated, and an invalid value returns 400.
4. Cache OpenRouterClient::listModels(). Right now it calls /models and /embeddings/models on every getConfiguredModels(), getMaxInputTokens() and getMaxOutputTokens() call.

Also found

• OpenRouterStreamedChatMessageIterator.php:35-36 reads completionTokenDetails, but the SDK property is completionTokensDetails. Because of this, streamed reasoning and cached token usage is always 0.
• OpenRouterClient builds the openai-php client without withHttpClient(), so ai.settings:request_timeout is not applied.

Remaining tasks

• MR
• Tests
• Review

User interface changes

A reasoning effort select appears in chat provider configuration for models that support it.

API changes

getAvailableConfiguration('chat', $model_id) returns model settings instead of an empty array.

Data model changes

None.