fix(e2e): Add per-flow retries to handle Tart VM timing flakiness

The warm-up flow helps but doesn't fully eliminate first-launch timing
issues on Cirrus Labs Tart VMs. On especially slow boots, the warm-up
itself can fail and the first few flows fail before the simulator
stabilises. Retry each flow up to 3 times to handle this.

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

Author: antonis

dev-packages/e2e-tests/cli.mjs

@@ -317,21 +317,31 @@ if (actions.includes('test')) {
     // Running all flows via `maestro test maestro` shares a single runner
     // session — if one flow (e.g. crash.yml) kills the app, subsequent flows
     // fail because Maestro's XCTest driver loses the connection.
+    // Retry failed flows up to 3 times — Tart VMs have transient first-launch
+    // timing issues where the app isn't ready when Maestro connects.
+    const maxAttempts = 3;
     const failed = [];
     try {
       for (const flowFile of flowFiles) {
-        try {
-          execFileSync('maestro', [
-            'test', `maestro/${flowFile}`, ...maestroEnvArgs,
-            '--debug-output', 'maestro-logs',
-            '--flatten-debug-output',
-          ], {
-            stdio: 'inherit',
-            cwd: e2eDir,
-          });
-        } catch (error) {
-          failed.push(flowFile);
+        let passed = false;
+        for (let attempt = 1; attempt <= maxAttempts && !passed; attempt++) {
+          try {
+            execFileSync('maestro', [
+              'test', `maestro/${flowFile}`, ...maestroEnvArgs,
+              '--debug-output', 'maestro-logs',
+              '--flatten-debug-output',
+            ], {
+              stdio: 'inherit',
+              cwd: e2eDir,
+            });
+            passed = true;
+          } catch (error) {
+            if (attempt < maxAttempts) {
+              console.warn(`Flow ${flowFile} failed (attempt ${attempt}/${maxAttempts}), retrying…`);
+            }
+          }
         }
+        if (!passed) failed.push(flowFile);
       }
     } finally {
       // Always redact sensitive data, even if the test fails