feat: 图片改图 MCP 工具 image_edit

- imagegen.edit：POST multipart /v1/images/edits，输入图
  必须是工作区内 plain basename 且存在（上限 20MB），产出
  必须是新文件、绝不覆盖输入；响应解析与 generate 共用
  _save_image_response（url 或 b64_json、下载落盘、格式探测）。
- imagegen_mcp 新增 image_edit 工具；image-workflow 技能
  改图分支从“无专用工具”改为用 image_edit。
- 测试：multipart 体含提示词与原图字节、输入缺失/同名/
  路径逃逸拒绝、MCP 列表含 image_edit；全量相关 16 项过。
实测依据（2026-10-30）：中转站图片分组 key 调
gpt-image-2.5-flare edits 返回 200，URL 图为真实 PNG
1024x1024、909K，usage 含输入图 token（1659 image tokens）。

cloud/imagegen_mcp.py

@@ -20,10 +20,10 @@ import os
 import sys
 
 try:
-    from cloud.imagegen import ImageGenError, generate
+    from cloud.imagegen import ImageGenError, edit, generate
 except ImportError:  # flat copy in a tenant codex-home
     sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
-    from imagegen import ImageGenError, generate  # noqa: E402
+    from imagegen import ImageGenError, edit, generate  # noqa: E402
 
 SERVER_NAME = "imagegen"
 SERVER_VERSION = "0.1.0"
@@ -61,6 +61,42 @@ _TOOLS = [
             "required": ["prompt"],
         },
     },
+    {
+        "name": "image_edit",
+        "description": (
+            "Edit an existing image in the current working directory "
+            "from a text prompt and save the result as a NEW file; "
+            "the input file is never overwritten. Returns the "
+            "absolute output path, format, dimensions when "
+            "detectable, byte size and the relay usage block. "
+            "Verify the returned path with shell/PIL before "
+            "reporting success."
+        ),
+        "inputSchema": {
+            "type": "object",
+            "properties": {
+                "prompt": {"type": "string",
+                           "description": "What to change, in detail"},
+                "input_filename": {"type": "string",
+                                   "description": "Plain basename "
+                                   "of the existing image in the "
+                                   "working directory"},
+                "output_filename": {"type": "string",
+                                    "description": "Plain basename "
+                                    "to save the edited image as. "
+                                    "Must differ from input_filename. "
+                                    "Default: generated-<epoch>.png"},
+                "size": {"type": "string", "default": "1024x1024",
+                         "description": "1024x1024, 1536x1024, "
+                         "1024x1536, or auto"},
+                "model": {"type": "string",
+                          "description": "Image model override; "
+                          "default is the IMAGE_GEN_MODEL env "
+                          "or gpt-image-2.5-flare"},
+            },
+            "required": ["prompt", "input_filename"],
+        },
+    },
 ]
 
 
@@ -68,11 +104,12 @@ def _err(code, message):
     return {"code": code, "message": message}
 
 
-def handle_request(request, generate_fn=None):
+def handle_request(request, generate_fn=None, edit_fn=None):
     """Handle one JSON-RPC request dict. Returns a response dict or
     None for notifications (no id). Pure function for testability:
-    the stdio loop passes the real generate, tests pass fakes."""
+    the stdio loop passes the real generate/edit, tests pass fakes."""
     generate_fn = generate_fn or generate
+    edit_fn = edit_fn or edit
     if not isinstance(request, dict):
         return {"jsonrpc": "2.0", "id": None,
                 "error": _err(-32700, "request is not an object")}
@@ -114,6 +151,13 @@ def handle_request(request, generate_fn=None):
                     output_filename=args.get("output_filename"),
                     size=args.get("size", "1024x1024"),
                     model=args.get("model"))
+            elif name == "image_edit":
+                result = edit_fn(
+                    args.get("prompt"),
+                    args.get("input_filename"),
+                    output_filename=args.get("output_filename"),
+                    size=args.get("size", "1024x1024"),
+                    model=args.get("model"))
             else:
                 return {"jsonrpc": "2.0", "id": req_id,
                         "error": _err(-32602, "unknown tool: %r" % name)}