feat: create embeddings tracing implementation (#1240)

**Description**

This adds embeddings tracing per OpenInference OpenAI semantics, and
adjusts docs accordingly.

**Related Issues/PRs**

Fixes #1085

---------

Signed-off-by: Adrian Cole <[email protected]>

Author: codefromthecrypt

cmd/aigw/README.md

@@ -51,7 +51,15 @@ Here are values we use for Ollama:
      docker compose run --rm embeddings
      ```
 
-4. **Shutdown the example stack**:
+4. **Create embeddings**:
+
+   The `create-embeddings` service uses `curl` to send an embeddings request
+   to the AI Gateway CLI (aigw) which routes it to Ollama.
+   ```bash
+   docker compose run --rm create-embeddings
+   ```
+
+5. **Shutdown the example stack**:
 
    `down` stops the containers and removes the volumes used by the stack.
    ```bash

internal/tracing/api/api.go

@@ -57,6 +57,11 @@ func (NoopTracing) ChatCompletionTracer() ChatCompletionTracer {
 	return NoopChatCompletionTracer{}
 }
 
+// EmbeddingsTracer implements Tracing.EmbeddingsTracer.
+func (NoopTracing) EmbeddingsTracer() EmbeddingsTracer {
+	return NoopEmbeddingsTracer{}
+}
+
 // ImageGenerationTracer implements Tracing.ImageGenerationTracer.
 func (NoopTracing) ImageGenerationTracer() ImageGenerationTracer {
 	return NoopImageGenerationTracer{}
@@ -142,6 +147,35 @@ func (NoopChatCompletionTracer) StartSpanAndInjectHeaders(context.Context, map[s
 	return nil
 }
 
+// EmbeddingsTracer creates spans for OpenAI embeddings requests.
+type EmbeddingsTracer interface {
+	// StartSpanAndInjectHeaders starts a span and injects trace context into
+	// the header mutation.
+	//
+	// Parameters:
+	//   - ctx: might include a parent span context.
+	//   - headers: Incoming HTTP headers used to extract parent trace context.
+	//   - headerMutation: The new Embeddings Span will have its context
+	//     written to these headers unless NoopTracing is used.
+	//   - req: The OpenAI embeddings request. Used to record request attributes.
+	//   - body: contains the original raw request body as a byte slice.
+	//
+	// Returns nil unless the span is sampled.
+	StartSpanAndInjectHeaders(ctx context.Context, headers map[string]string, headerMutation *extprocv3.HeaderMutation, req *openai.EmbeddingRequest, body []byte) EmbeddingsSpan
+}
+
+// EmbeddingsSpan represents an OpenAI embeddings request.
+type EmbeddingsSpan interface {
+	// RecordResponse records the response attributes to the span.
+	RecordResponse(resp *openai.EmbeddingResponse)
+
+	// EndSpanOnError finalizes and ends the span with an error status.
+	EndSpanOnError(statusCode int, body []byte)
+
+	// EndSpan finalizes and ends the span.
+	EndSpan()
+}
+
 // ImageGenerationTracer creates spans for OpenAI image generation requests.
 type ImageGenerationTracer interface {
 	// StartSpanAndInjectHeaders starts a span and injects trace context into

internal/tracing/tracing.go

@@ -39,16 +39,16 @@ func (t *tracingImpl) ChatCompletionTracer() tracing.ChatCompletionTracer {
 	return t.chatCompletionTracer
 }
 
-// ImageGenerationTracer implements the same method as documented on api.Tracing.
-func (t *tracingImpl) ImageGenerationTracer() tracing.ImageGenerationTracer {
-	return t.imageGenerationTracer
-}
-
 // EmbeddingsTracer implements the same method as documented on api.Tracing.
 func (t *tracingImpl) EmbeddingsTracer() tracing.EmbeddingsTracer {
 	return t.embeddingsTracer
 }
 
+// ImageGenerationTracer implements the same method as documented on api.Tracing.
+func (t *tracingImpl) ImageGenerationTracer() tracing.ImageGenerationTracer {
+	return t.imageGenerationTracer
+}
+
 func (t *tracingImpl) MCPTracer() tracing.MCPTracer {
 	return t.mcpTracer
 }