fix(e2e): raise live provision/claim-deploy timeout 120→180s (de-flake)

e2e/live-anon-provision.spec.ts

@@ -255,6 +255,15 @@ test.describe('LIVE — every anonymous provision flow → backend-assert → re
     }: {
       request: APIRequestContext
     }) => {
+      // Real prod provisioning (esp. /vector = CREATE DATABASE + CREATE
+      // EXTENSION, /nosql, /db) + the assert-usable connect from the CI runner
+      // occasionally spikes well past Playwright's 120s default under prod
+      // contention — normally ~14s, but a busy provisioner has been seen to
+      // exceed 2min, timing out the whole serial group (one slow flow aborts the
+      // rest). Give the live provision+connect+reap real headroom so transient
+      // latency flakes (recovered on retry today) stop redding the suite. This
+      // verifies the flow WORKS, not that it's <120s.
+      test.setTimeout(180_000)
       const name = cohortName(`anon-${flow.label}`)
 
       // ── Create: real provision against the live api ────────────────────────

e2e/live-claim-deploy.spec.ts

@@ -232,6 +232,15 @@ async function provisionInEnv(
 test.describe('LIVE — W3 claim/conversion + deploy-lifecycle + env-switcher (cross-surface)', () => {
   test.describe.configure({ mode: 'serial' })
 
+  // These legs do real prod provisioning + claim + deploy (each heavy) and the
+  // assert-usable connect from the CI runner — occasionally well past the 120s
+  // default under prod contention, timing out the serial group on transient
+  // latency (recovered on retry). Raise the per-test budget so a slow-but-OK
+  // run doesn't red the suite. See live-anon-provision.spec.ts for the same.
+  test.beforeEach(() => {
+    test.setTimeout(180_000)
+  })
+
   // Hard skip in normal CI: the LIVE harness must never make the per-PR gate
   // depend on a reachable backend.
   test.skip(