Poll CI every 100ms in the test harness

tests/ci.rs

@@ -488,9 +488,9 @@ fn waits_for_github_to_work_out_whether_the_pr_is_mergeable() {
         checks_on_head(GREEN)
     ));
 
-    // Each poll starts the fake gh, so the default 300ms grace period can run
-    // out before the fourth read on a slow machine. Green checks keep the
-    // longer grace period from slowing the CI wait.
+    // At the harness's 100ms poll interval, the default 300ms grace period
+    // holds only about three reads, and this needs a fourth. Green checks keep
+    // the longer grace period from slowing the CI wait.
     let result = scenario.run_with_env(
         &[&scenario.issue_url(7)],
         &[("THIRDSHIFT_CI_GRACE_MS", "5000")],

tests/support/mod.rs

@@ -388,9 +388,11 @@ impl Scenario {
             .env("FAKE_CLAUDE_SCRIPT", self.path("claude-script.sh"))
             .env("FAKE_CLAUDE_RECORD", self.path("claude-calls.json"))
             .env("FAKE_GH_RECORD", self.path("gh-calls.json"))
-            // Seconds of waiting for CI become milliseconds.
+            // Seconds of waiting for CI become milliseconds. Each poll starts
+            // the fake gh, whose Python startup costs CPU; at 100ms the grace
+            // period holds about three reads, enough for the absent state.
             .env("THIRDSHIFT_CI_GRACE_MS", "300")
-            .env("THIRDSHIFT_POLL_MS", "10");
+            .env("THIRDSHIFT_POLL_MS", "100");
         command
     }