Enhance backend loading diagnostics and remount logic

- Introduced a new `loading_fail_sec` parameter in the `wait_backend_idle` function to handle prolonged loading states, improving error handling for backend readiness.
- Updated the `ensure_dlbackend_bind` function to stop SwarmUI before remounting, preventing target busy errors and ensuring consistent data mounts.
- Enhanced the `recover_errored_backends` function to account for the new remount logic, improving backend recovery processes.
- Refactored tests to validate the new loading failure conditions and ensure proper handling of backend states during diagnostics.

src/gpu_rent/ready.py

@@ -208,17 +208,21 @@ def wait_backend_idle(
     poll_every: float = 15.0,
     errored_fail_sec: float = 120.0,
     disabled_fail_sec: float = 90.0,
+    loading_fail_sec: float = 900.0,
 ) -> None:
     """Block until SwarmUI backends are ready (status=running, no queue).
 
     Note: SwarmUI ``idle`` means suspended backends (cannot generate).
-    Ready-to-use is ``running``. Sustained ``errored`` / ``disabled`` fail-fasts.
+    Ready-to-use is ``running``. Sustained ``errored`` / ``disabled`` / long
+    ``loading`` fail-fasts.
     """
     deadline = time.time() + timeout
     log("жду ready backend (running)…")
     last = ""
     errored_since: float | None = None
     disabled_since: float | None = None
+    loading_since: float | None = None
+    last_loading_note = 0.0
     pretty = {
         "BUSY backend=loading (Comfy стартует)": "… Comfy стартует",
         "BUSY backend=some_loading (Comfy стартует)": "… Comfy стартует (часть бэкендов)",
@@ -242,6 +246,38 @@ def wait_backend_idle(
             out = f"WAIT ssh: {exc}"
         line = out.splitlines()[-1] if out else "WAIT empty"
         shown = pretty.get(line, line)
+        if "BUSY backend=loading" in line or "BUSY backend=some_loading" in line:
+            now = time.time()
+            if loading_since is None:
+                loading_since = now
+            elapsed = int(now - loading_since)
+            shown = f"… Comfy стартует ({elapsed // 60}м {elapsed % 60}с)"
+            if now - last_loading_note >= 60:
+                last_loading_note = now
+                try:
+                    j = run_ssh(
+                        cfg,
+                        host,
+                        "sudo -n journalctl -u swarmui -n 8 --no-pager -o cat 2>/dev/null "
+                        "| tail -n 8 || true",
+                        check=False,
+                        timeout=25,
+                    ).strip()
+                    if j:
+                        for jl in j.splitlines()[-4:]:
+                            log(f"  journal: {jl[:160]}")
+                except Exception:
+                    pass
+            if elapsed >= loading_fail_sec:
+                collect_swarm_diagnostics(cfg, host, log)
+                raise CloudError(
+                    f"backend=loading уже {elapsed} с — похоже завис "
+                    "(часто FrontendVersion / pip / сеть). Диагностика выше. "
+                    "Попробуй gpu-rent up снова или Server → Backends → Restart."
+                )
+        else:
+            loading_since = None
+
         if shown != last:
             log(shown)
             last = shown