Route xAI BYOK models through correct default prompts

[pwang347]

Validated that the behaviour change does not seem drastic. Given relatively lower usage from BYOK route, I think we can make all the prompts consistent for xAI models without need for ExP.

Note: ideally we should pass the provider's byok ID directly, but as is there's a layering issue.

[Copilot]

Pull Request Overview

This PR adds a matchesModel static method to the XAIPromptResolver class to enable more precise model matching for xAI endpoints. This allows the xAI prompt resolver to be matched based on the endpoint's family property ("xAI") rather than relying solely on the model name prefix ("grok-code").

Key Changes

• Added a matchesModel static method to XAIPromptResolver that checks if endpoint.family === 'xAI'

[Copilot]

The matchesModel implementation checks for endpoint.family === 'xAI', while familyPrefixes is set to ['grok-code']. This creates a discrepancy where the matcher uses the family property ('xAI') but the prefix fallback uses a different value ('grok-code'). According to the PromptRegistry.getPrompt logic, matchers are checked first, so the prefix list may never be used for xAI endpoints. Consider either aligning the familyPrefixes with the actual family name (e.g., ['xAI']) or documenting why these values differ.