fix: skip verify_apps, close_all, activate_window in lightweight mode (#177)

These setup entry types hang (120s timeout), crash the WAA server, or
are unnecessary for task execution. In lightweight mode (the default),
they are now skipped entirely — both the verify_apps step injected from
related_apps and any close_all / activate_window entries in the task
config array. Each skipped entry is recorded with status "skipped" in
_last_setup_results for auditability.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: abrichr

openadapt_evals/adapters/waa/live.py

@@ -1928,7 +1928,7 @@ def _run_task_setup(self, raw_config: dict) -> None:
         self._last_setup_results = []
 
         related_apps = raw_config.get("related_apps", [])
-        if related_apps:
+        if related_apps and not self.config.lightweight:
             verify_step = {
                 "type": "verify_apps",
                 "parameters": {"apps": related_apps},
@@ -1944,6 +1944,19 @@ def _run_task_setup(self, raw_config: dict) -> None:
         for entry in config:
             entry_type = entry.get("type", "unknown")
 
+            # In lightweight mode, skip non-critical setup entries that are
+            # known to hang or crash the WAA server.
+            if self.config.lightweight and entry_type in (
+                "verify_apps",
+                "close_all",
+                "activate_window",
+            ):
+                logger.debug("Lightweight mode: skipping %s", entry_type)
+                self._last_setup_results.append(
+                    {"type": entry_type, "status": "skipped"}
+                )
+                continue
+
             # Sleep is handled locally, not on the server
             if entry_type == "sleep":
                 duration = entry.get("parameters", {}).get("seconds", 1)