fix(winrm): retry transient WSMan host-process faults (1018)

After a memory-heavy parallel build the guest can momentarily fail to
launch wsmprovhost (WSManFault 1018), which broke artifact collect
(now does Compress-Archive + fetch over WinRM). run() retries up to 4x
with exponential backoff on transient WSMan markers, dropping the
cached shell so a fresh one is opened once the guest recovers.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/ci_orchestrator/transport/winrm.py

@@ -9,12 +9,24 @@ for an isolated lab network (see ``scripts/_Common.psm1``).
 from __future__ import annotations
 
 import contextlib
+import time
 from dataclasses import dataclass
 from pathlib import Path
 from typing import TYPE_CHECKING, Any
 
 from ci_orchestrator.transport.errors import TransportCommandError, TransportConnectError
 
+# WSMan faults that are transient — the guest momentarily cannot launch
+# the WSMan host process (wsmprovhost), typically right after a
+# memory-heavy build. Retrying after a short settle usually succeeds.
+_WSMAN_TRANSIENT_MARKERS = (
+    "1018",  # could not launch a host process to process the given request
+    "could not launch a host process",
+    "the wsman provider host server",
+    "1726",  # RPC server unavailable (service still coming back)
+    "connection was aborted",
+)
+
 if TYPE_CHECKING:  # pragma: no cover
     from pypsrp.client import Client
 
@@ -118,12 +130,37 @@ class WinRmTransport:
 
     # --------------------------------------------------------------- ops
     def run(self, script: str, *, check: bool = True) -> WinRmResult:
-        """Run a PowerShell script block in the guest."""
+        """Run a PowerShell script block in the guest.
-        client = self._connect()
+
-        try:
+        Retries transient WSMan host-process-launch failures (e.g. fault
-            stdout, stderr, returncode = client.execute_ps(script)
+        1018 right after a memory-heavy build) with a short backoff.
-        except Exception as exc:
+        """
-            raise TransportConnectError(f"WinRM execute_ps failed: {exc}") from exc
+        attempts = 4
+        delay = 5.0
+        last_exc: Exception | None = None
+        for attempt in range(attempts):
+            client = self._connect()
+            try:
+                stdout, stderr, returncode = client.execute_ps(script)
+                break
+            except Exception as exc:
+                last_exc = exc
+                msg = str(exc).lower()
+                transient = any(m in msg for m in _WSMAN_TRANSIENT_MARKERS)
+                if transient and attempt < attempts - 1:
+                    # Drop the cached client so the next try opens a fresh
+                    # WSMan shell once the guest can spawn the host again.
+                    self.close()
+                    time.sleep(delay)
+                    delay *= 2
+                    continue
+                raise TransportConnectError(
+                    f"WinRM execute_ps failed: {exc}"
+                ) from exc
+        else:  # pragma: no cover - loop always breaks or raises
+            raise TransportConnectError(
+                f"WinRM execute_ps failed: {last_exc}"
+            ) from last_exc
         # pypsrp returns stderr as a PSDataStreams object (newer versions) or a
         # list of error records (older versions); normalise to a plain string.
         stderr_any: Any = stderr