Add timeout params for PUT inference and POST inference/_stream

Author: davidkyle

output/openapi/elasticsearch-openapi.json

@@ -21,6 +21,7 @@ import { InferenceEndpoint } from '@inference/_types/Services'
 import { TaskType } from '@inference/_types/TaskType'
 import { RequestBase } from '@_types/Base'
 import { Id } from '@_types/common'
+import { Duration } from '@_types/Time'
 
 /**
  * Create an inference endpoint.
@@ -60,6 +61,13 @@ export interface Request extends RequestBase {
      */
     inference_id: Id
   }
+  query_parameters: {
+    /**
+     * Specifies the amount of time to wait for the inference endpoint to be created.
+     * @server_default 30s
+     */
+    timeout?: Duration
+  }
   /** @codegen_name inference_config */
   body: InferenceEndpoint
 }

output/openapi/elasticsearch-serverless-openapi.json

@@ -20,6 +20,7 @@
 import { TaskType } from '@inference/_types/TaskType'
 import { RequestBase } from '@_types/Base'
 import { Id } from '@_types/common'
+import { Duration } from '@_types/Time'
 
 /**
  * Perform streaming inference.
@@ -55,6 +56,13 @@ export interface Request extends RequestBase {
      */
     task_type?: TaskType
   }
+  query_parameters: {
+    /**
+     * The amount of time to wait for the inference request to complete.
+     * @server_default 30s
+     */
+    timeout?: Duration
+  }
   body: {
     /**
      * The text on which you want to perform the inference task.