more optimize of ai search plugin (#1896)

plugins/wasm-go/extensions/ai-search/README_EN.md

@@ -20,6 +20,7 @@ Plugin execution priority: `440`
 | defaultEnable | bool | Optional | true | Whether the plugin functionality is enabled by default. When set to false, the plugin will only be activated when the request contains a web_search_options field |
 | needReference | bool | Optional | false | Whether to add reference sources in the response |
 | referenceFormat | string | Optional | `"**References:**\n%s"` | Reference content format, must include %s placeholder |
+| referenceLocation | string | Optional | "head" | Reference position: "head" at the beginning of the response, "tail" at the end of the response |
 | defaultLang | string | Optional | - | Default search language code (e.g. zh-CN/en-US) |
 | promptTemplate | string | Optional | Built-in template | Prompt template, must include `{search_results}` and `{question}` placeholders |
 | searchFrom | array of object | Required | - | Refer to search engine configuration below, at least one engine must be configured |
@@ -45,6 +46,7 @@ It is strongly recommended to enable this feature when using Arxiv or Elasticsea
 | llmUrl | string | Required | - | LLM service API URL |
 | llmModelName | string | Required | - | LLM model name |
 | timeoutMillisecond | number | Optional | 30000 | API call timeout (milliseconds) |
+| maxCount | number | Optional | 3 | Maximum number of search queries generated by the search rewrite |
 
 ## Search Engine Common Configuration
 
@@ -224,6 +226,18 @@ searchFrom:
   servicePort: 8080
 ```
 
+### Custom Reference Location
+
+```yaml
+needReference: true
+referenceLocation: "tail"  # Add references at the end of the response instead of the beginning
+searchFrom:
+- type: bing
+  apiKey: "your-bing-key"
+  serviceName: "search-service.dns"
+  servicePort: 8080
+```
+
 ### Search Rewrite Configuration
 
 ```yaml
@@ -258,6 +272,25 @@ searchFrom:
 
 This configuration is suitable for models that support web search options, such as OpenAI's gpt-4o-search-preview model. When the request contains a `web_search_options` field, even if it's an empty object (`"web_search_options": {}`), the plugin will be activated.
 
+### Search Context Size Configuration
+
+You can dynamically adjust the number of search queries by adding a `search_context_size` parameter in the `web_search_options` field of the request:
+
+```json
+{
+  "web_search_options": {
+    "search_context_size": "medium"
+  }
+}
+```
+
+The `search_context_size` supports three levels:
+- `low`: Generates 1 search query (suitable for simple questions)
+- `medium`: Generates 3 search queries (default)
+- `high`: Generates 5 search queries (suitable for complex questions)
+
+This setting overrides the `maxCount` value in the configuration, allowing clients to dynamically adjust search depth based on question complexity.
+
 ## Notes
 
 1. The prompt template must include `{search_results}` and `{question}` placeholders, optionally use `{cur_date}` to insert current date (format: January 2, 2006)