Fix OpenAI transcription zero token usage

[Gujiassh]

Summary

• preserve explicit zero-valued transcription token usage fields instead of falling back to prompt/completion token keys
• add focused provider coverage for both the zero-valued and fallback usage paths

Testing

• bundle exec rspec spec/ruby_llm/providers/open_ai/transcription_spec.rb
• bundle exec rubocop lib/ruby_llm/providers/openai/transcription.rb spec/ruby_llm/providers/open_ai/transcription_spec.rb

[Gujiassh]

Refreshed this branch onto the latest main and reverified the merged branch tip locally.

Local verification after the refresh:

• bundle exec rspec spec/ruby_llm/providers/open_ai/transcription_spec.rb spec/ruby_llm/generators/chat_ui_generator_spec.rb → passed (22 examples, 0 failures)
• bundle exec rubocop lib/ruby_llm/providers/openai/transcription.rb spec/ruby_llm/providers/open_ai/transcription_spec.rb spec/ruby_llm/generators/chat_ui_generator_spec.rb → passed

The zero-token OpenAI transcription handling is still intact on the refreshed branch.

[Gujiassh]

Closing this stale radar PR to keep the queue clean: it has been open for weeks without maintainer response, and I am not going to keep it in the active follow-up set without renewed upstream interest.