feat: implement dynamic model cost resolver via models.dev API

[ganimalqudhaifi]

This PR introduces a dynamic pricing fetcher that automatically retrieves the latest model costs from the models.dev API registry.

Key Improvements:

• Zero Hardcoding: Eliminates the need to manually hardcode and update prices for Google, Anthropic, and OpenAI models inside the plugin.
• Dynamic Provider Mapping: Automatically maps OpenCode model IDs to their respective providers in the API registry.
• Smart Caching: Responses are securely cached in the OS tmpdir with a 24-hour TTL to ensure CLI boot times remain instant without unnecessary network overhead.
• Conflict-Free: Designed to be completely modular and will seamlessly merge alongside other pending PRs (e.g., model capabilities alignment).

[anthonyhaussman]

Thorough code review performed. Several critical and high severity issues were found related to blocking network requests on CLI boot and shared cache permissions. Please see the line-specific comments for detailed feedback.