OpenAI prompt caching: parse and report cached token usage

[bug-ops]

Summary

• Parse cached_tokens from OpenAI response usage.prompt_tokens_details
• Add Mutex<Option<(u64, u64)>> cache field to OpenAiProvider (mirrors Claude pattern)
• Implement last_cache_usage() returning (0, cached_tokens) for MetricsCollector
• 7 unit tests covering deserialization, storage, edge cases, and clone behavior

References

Closes #348, closes #349, closes #350, closes #351

Test plan

• cargo nextest run --workspace --lib --bins -- 1416 passed
• cargo clippy --workspace -- -D warnings -- zero warnings
• Performance validation -- minimal Mutex scope, no allocations in hot path
• Security validation -- no sensitive data in logs or debug output

[codecov-commenter]

Codecov Report

❌ Patch coverage is 93.93939% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/zeph-llm/src/openai.rs	93.93%	6 Missing ⚠️

@@            Coverage Diff             @@
##             main     #352      +/-   ##
==========================================
+ Coverage   79.33%   79.38%   +0.05%     
==========================================
  Files         100      100              
  Lines       25665    25764      +99     
==========================================
+ Hits        20361    20454      +93     
- Misses       5304     5310       +6     

Files with missing lines	Coverage Δ
crates/zeph-llm/src/openai.rs	76.73% <93.93%> (+2.07%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.