feat(model-manager): "Free GPU memory" button to unload ComfyUI models

ComfyUI caches the last model when RAM is plentiful (unified memory), so
memory doesn't drop after switching models even though models are being
swapped, not accumulated. Add a sidebar "Free GPU memory" button that
proxies ComfyUI's POST /free (unload_models + free_memory) via a new
/api/comfyui/free endpoint (COMFYUI_URL env). Verified it releases ~7GB.
README documents this plus the --disable-smart-memory auto-unload option.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

model-manager/app/main.py

@@ -15,6 +15,7 @@ from pydantic import BaseModel
 from . import db, downloader, registries
 from .config import (
     COMFYUI_PORT,
+    COMFYUI_URL,
     COMFYUIMINI_PORT,
     IMAGE_EXTS,
     KEY_BY_FOLDER,
@@ -314,6 +315,20 @@ def ui_config() -> dict:
     return {"comfyui_port": COMFYUI_PORT, "comfyuimini_port": COMFYUIMINI_PORT}
 
 
+@app.post("/api/comfyui/free")
+async def comfyui_free() -> dict:
+    """Ask ComfyUI to unload all models and free GPU/RAM (proxies POST /free)."""
+    try:
+        async with httpx.AsyncClient(timeout=30) as client:
+            resp = await client.post(
+                f"{COMFYUI_URL}/free",
+                json={"unload_models": True, "free_memory": True})
+            resp.raise_for_status()
+    except Exception as exc:  # noqa: BLE001 - report to UI
+        raise HTTPException(502, f"Could not reach ComfyUI: {exc}")
+    return {"ok": True}
+
+
 @app.get("/start")
 def start() -> FileResponse:
     return FileResponse(STATIC_DIR / "start.html")