Add instrumentation for Batch API

The OpenAI integration does not instrument the Batch API. Batch API usage is growing for cost-optimized bulk inference.

**Endpoints to instrument:**

* `openai.resources.Batches.create()` (sync + async)
* `openai.resources.Batches.retrieve()` (sync + async)
* `openai.resources.Batches.list()` (sync + async)
* `openai.resources.Batches.cancel()` (sync + async)

**Attributes to capture:**

* Batch ID
* Target endpoint (e.g. `/v1/chat/completions`)
* Batch status (validating, in_progress, completed, failed, etc.)
* Token usage from completed batches (input, output, total)
* Request/completion counts
* Follow OTel GenAI semantic conventions for span naming and attributes

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