Add instrumentation for Audio endpoints (transcriptions, translations, speech)

The OpenAI integration does not instrument any Audio API endpoints.

**Endpoints to instrument:**

* `openai.resources.Audio.transcriptions.create()` — speech-to-text (sync + async)
* `openai.resources.Audio.translations.create()` — audio translation (sync + async)
* `openai.resources.Audio.speech.create()` — text-to-speech (sync + async)

**Attributes to capture:**

* Model name
* Input text / audio metadata (if PII settings allow)
* Language, response format, temperature parameters
* Duration and token usage where applicable
* Follow OTel GenAI semantic conventions for span naming and attributes

**Reference:** [OpenAI integration compliance review](<../openai-integration-compliance-review.md>), Section 6