feat: map LiteLLM model prices into OpenCode cost field (#22)

Populate Model.cost from /v1/model/info pricing metadata instead of hardcoding zero. This makes OpenCode's spent/cost tracking match the actual LiteLLM proxy billing.

Also adds cache_read_input_token_cost and cache_creation_input_token_cost to LiteLLMModelInfo so Anthropic-style prompt caching prices are propagated.

Co-authored-by: Contributor <noreply@github.com>