Implement last_cache_usage() and add unit tests for OpenAI cache tracking

[bug-ops]

Parent: #348
Depends on: #349, #350

Task

Implement the last_cache_usage() trait method and add comprehensive tests.

Implementation

1. Add last_cache_usage() override in impl LlmProvider for OpenAiProvider:

fn last_cache_usage(&self) -> Option<(u64, u64)> {
    self.last_cache.lock().ok().and_then(|g| *g)
}

2. Add unit tests:

• OpenAiUsage deserialization with and without prompt_tokens_details
• OpenAiUsage deserialization with cached_tokens: 0
• PromptTokensDetails deserialization
• OpenAiChatResponse with usage field present
• ToolChatResponse with usage field present
• last_cache_usage() returns None initially
• Verify (0, cached_tokens) convention

Acceptance criteria

• last_cache_usage() returns correct values after API call simulation
• All new tests pass
• Existing tests unaffected
• cargo clippy --workspace -- -D warnings clean