feat: Add optional idempotency support to batches API

Implements optional idempotency for batch creation using `idem_tok` parameter:

* **Core idempotency**: Same token + parameters returns existing batch
* **Conflict detection**: Same token + different parameters raises HTTP 409 ConflictError
* **Metadata order independence**: Different key ordering doesn't affect idempotency

**API changes:**
- Add optional `idem_tok` parameter to `create_batch()` method
- Enhanced API documentation with idempotency extensions

**Implementation:**
- Reference provider supports idempotent batch creation
- ConflictError for proper HTTP 409 status code mapping
- Comprehensive parameter validation

**Testing:**
- Unit tests: focused tests covering core scenarios with parametrized conflict detection
- Integration tests: tests validating real OpenAI client behavior

This enables client-side retry safety and prevents duplicate batch creation
when using the same idempotency token, following REST API

llama_stack/apis/batches/batches.py

@@ -29,12 +29,16 @@ class ListBatchesResponse(BaseModel):
 
 @runtime_checkable
 class Batches(Protocol):
-    """Protocol for batch processing API operations.
-
+    """
     The Batches API enables efficient processing of multiple requests in a single operation,
     particularly useful for processing large datasets, batch evaluation workflows, and
     cost-effective inference at scale.
 
+    The API is designed to allow use of openai client libraries for seamless integration.
+
+    This API provides the following extensions:
+     - idempotent batch creation
+
     Note: This API is currently under active development and may undergo changes.
     """
 
@@ -45,6 +49,7 @@ class Batches(Protocol):
         endpoint: str,
         completion_window: Literal["24h"],
         metadata: dict[str, str] | None = None,
+        idem_tok: str | None = None,  # intentionally bad name
     ) -> BatchObject:
         """Create a new batch for processing multiple API requests.
 
@@ -52,6 +57,7 @@ class Batches(Protocol):
         :param endpoint: The endpoint to be used for all requests in the batch.
         :param completion_window: The time window within which the batch should be processed.
         :param metadata: Optional metadata for the batch.
+        :param idem_tok: Optional idempotency token. When provided, enables idempotent behavior.
         :returns: The created batch object.
         """
         ...