Implement idle-killer enhancements and swarm management improvements

- Marked critical bugs as resolved in the review documentation, including changes to the `arm_idle_killer` function to raise errors on credential creation failures and ensure proper file permissions for JSON credentials.
- Introduced a new `_try_arm_idle_killer` function in `provision.py` to manage idle-killer state more effectively, ensuring it arms correctly during provisioning.
- Updated the `swarm_busy` function in `remote/idle_killer.py` to allow idle state after a specified duration of Swarm unavailability, preventing unnecessary billing.
- Enhanced performance tuning logic in `tune_swarm_perf.py` to ensure proper handling of pip installation success before applying extra arguments.
- Added tests to validate the new idle-killer behavior and swarm management logic, ensuring robustness in handling idle states and error conditions.

src/gpu_rent/remote/idle_killer.py

@@ -13,9 +13,12 @@ DATA = Path("/mnt/swarm_data")
 CREDS = Path("/root/.gpu-rent/idle-killer.json")
 HOLD = DATA / ".gpu-rent-hold-until"
 IDLE_SINCE = DATA / ".gpu-rent-idle-since"
+SWARM_DOWN_SINCE = DATA / ".gpu-rent-swarm-down-since"
 ARMED = DATA / ".gpu-rent-killer-armed"
 LOG = DATA / ".gpu-rent-killer.log"
 SERVER_ID_FILE = DATA / ".gpu-rent-server-id"
+# After this many seconds of continuous Swarm unreachable → treat as idle (don't bill forever).
+SWARM_UNREACHABLE_IDLE_SEC = 60 * 60
 
 
 def log(msg: str) -> None:
@@ -49,7 +52,11 @@ def write_ts(path: Path, value: float | None) -> None:
 
 
 def swarm_busy(swarm_url: str, timeout: float = 8.0) -> tuple[bool, str]:
-    """Return (busy, detail). Treat unreachable UI as busy (don't kill mid-boot)."""
+    """Return (busy, detail).
+
+    Unreachable UI is busy during boot, but after SWARM_UNREACHABLE_IDLE_SEC of
+    continuous failure we stop counting it as busy so idle clock can run / delete.
+    """
     ctx = ssl.create_default_context()
     try:
         req = urllib.request.Request(
@@ -73,7 +80,18 @@ def swarm_busy(swarm_url: str, timeout: float = 8.0) -> tuple[bool, str]:
         with urllib.request.urlopen(req2, timeout=timeout, context=ctx) as resp:
             data = json.loads(resp.read().decode("utf-8"))
     except (urllib.error.URLError, urllib.error.HTTPError, TimeoutError, json.JSONDecodeError, OSError) as exc:
-        return True, f"swarm unreachable: {exc}"
+        now = time.time()
+        since = read_ts(SWARM_DOWN_SINCE)
+        if since is None:
+            write_ts(SWARM_DOWN_SINCE, now)
+            return True, f"swarm unreachable (clock start): {exc}"
+        down_for = now - since
+        if down_for >= SWARM_UNREACHABLE_IDLE_SEC:
+            return False, f"swarm unreachable {int(down_for)}s >= {SWARM_UNREACHABLE_IDLE_SEC}s — allow idle"
+        return True, f"swarm unreachable {int(down_for)}s / {SWARM_UNREACHABLE_IDLE_SEC}s: {exc}"
+
+    # Reachable again — clear down clock.
+    write_ts(SWARM_DOWN_SINCE, None)
 
     status = data.get("status") or {}
     backend = data.get("backend_status") or {}