feat(server): add passthrough proxy for OpenAI-compatible endpoints

* implement proxy handler for various OpenAI API routes
* add error handling for request body and response streaming
* introduce new error response format for API compatibility
* add tests for recover middleware to handle panics gracefully

pkg/server/spec/openapi.yaml

@@ -1,7 +1,11 @@
 openapi: "3.1.0"
 info:
   title: vecna Embedding Adapter
-  description: Proxies text to a backing embedding model and adapts the result vectors between dimensions.
+  description: >
+    Proxies text to a backing embedding model and adapts the result vectors
+    between dimensions. Also forwards the rest of the OpenAI-compatible API
+    surface (chat, completions, models, audio, images, moderations, rerank)
+    verbatim to the resolved target, with no adaptation applied.
   version: "1.0.0"
 
 servers:
@@ -21,7 +25,14 @@ components:
       type: object
       properties:
         error:
-          type: string
+          type: object
+          properties:
+            message:
+              type: string
+            type:
+              type: string
+            code:
+              type: integer
 
     OpenAIEmbedRequest:
       type: object
@@ -169,6 +180,40 @@ paths:
         "502":
           description: Backing model error
 
+    get:
+      summary: OpenAI-compatible embeddings (query-param convenience form)
+      operationId: openaiEmbeddingsGet
+      parameters:
+        - name: input
+          in: query
+          required: true
+          description: Repeatable — one or more texts to embed.
+          schema:
+            type: array
+            items:
+              type: string
+        - name: model
+          in: query
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Adapted embeddings
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/OpenAIEmbedResponse'
+        "400":
+          description: Bad request
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/Error'
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
   /v1/models/{model}:embedContent:
     post:
       summary: Google-compatible single embedContent
@@ -250,3 +295,158 @@ paths:
           description: Unauthorized
         "502":
           description: Backing model error
+
+  # --- Generic OpenAI-compatible passthrough -------------------------------
+  # Forwarded verbatim to the resolved target (chosen by the body's "model"
+  # field, falling back to forward.default). No vecna-specific processing —
+  # request/response shape is whatever the backing model's API defines.
+  # Each also exists under /map/{mapping}/... to force a specific target.
+  /v1/models:
+    get:
+      summary: List models (passthrough)
+      operationId: listModels
+      responses:
+        "200":
+          description: Backend-defined model list
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/models/{model}:
+    get:
+      summary: Retrieve a model (passthrough)
+      operationId: retrieveModel
+      parameters:
+        - name: model
+          in: path
+          required: true
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Backend-defined model object
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/chat/completions:
+    post:
+      summary: Chat completions (passthrough, streaming supported)
+      operationId: chatCompletions
+      responses:
+        "200":
+          description: Backend-defined chat completion (or SSE stream if "stream":true)
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/completions:
+    post:
+      summary: Legacy completions (passthrough, streaming supported)
+      operationId: completions
+      responses:
+        "200":
+          description: Backend-defined completion (or SSE stream if "stream":true)
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/moderations:
+    post:
+      summary: Moderations (passthrough)
+      operationId: moderations
+      responses:
+        "200":
+          description: Backend-defined moderation result
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/images/generations:
+    post:
+      summary: Image generation (passthrough)
+      operationId: imageGenerations
+      responses:
+        "200":
+          description: Backend-defined image result
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/images/edits:
+    post:
+      summary: Image edits (passthrough, multipart/form-data)
+      operationId: imageEdits
+      responses:
+        "200":
+          description: Backend-defined image result
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/images/variations:
+    post:
+      summary: Image variations (passthrough, multipart/form-data)
+      operationId: imageVariations
+      responses:
+        "200":
+          description: Backend-defined image result
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/audio/speech:
+    post:
+      summary: Text-to-speech (passthrough)
+      operationId: audioSpeech
+      responses:
+        "200":
+          description: Backend-defined audio bytes
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/audio/transcriptions:
+    post:
+      summary: Speech-to-text (passthrough, multipart/form-data)
+      operationId: audioTranscriptions
+      responses:
+        "200":
+          description: Backend-defined transcription
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/audio/translations:
+    post:
+      summary: Speech translation (passthrough, multipart/form-data)
+      operationId: audioTranslations
+      responses:
+        "200":
+          description: Backend-defined translation
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error
+
+  /v1/rerank:
+    post:
+      summary: Rerank (passthrough — Cohere/Infinity/vLLM-style)
+      operationId: rerank
+      responses:
+        "200":
+          description: Backend-defined rerank result
+        "401":
+          description: Unauthorized
+        "502":
+          description: Backing model error