Pin backend test lanes to ubuntu-24.04

.github/workflows/backend-e2e.yml

@@ -62,7 +62,8 @@ env:
 jobs:
   e2e-http:
     name: dotnet test (E2ETests · HTTP · Postgres)
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     timeout-minutes: 30
 
     services:
@@ -244,7 +245,8 @@ jobs:
 
   e2e-worker:
     name: recurring jobs fire (worker host · Postgres)
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     timeout-minutes: 20
 
     services:

.github/workflows/backend-integration.yml

@@ -37,7 +37,8 @@ env:
 jobs:
   integration:
     name: dotnet test (IntegrationTests · Postgres)
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     # The suite itself takes ~25 min on a hosted runner, so 30 left no headroom: a slow runner was
     # cancelled mid-run and reported as a cancellation, not a failure. 45 keeps a real hang bounded
     # (--blame-hang-timeout below still dumps and aborts a single stuck test after 5 min).

.github/workflows/backend-unit.yml

@@ -41,7 +41,8 @@ env:
 jobs:
   unit:
     name: dotnet test (UnitTests)
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     timeout-minutes: 20
     steps:
       - uses: actions/checkout@v4

.github/workflows/flake-hunt.yml

@@ -40,7 +40,8 @@ jobs:
   unit:
     name: repeat (UnitTests)
     if: inputs.project == 'unit'
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     timeout-minutes: 180
     steps:
       - uses: actions/checkout@v4
@@ -108,7 +109,8 @@ jobs:
   integration:
     name: repeat (IntegrationTests · Postgres)
     if: inputs.project == 'integration'
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     timeout-minutes: 180
 
     env:

.github/workflows/large-capture.yml

@@ -51,7 +51,8 @@ env:
 jobs:
   large-capture:
     name: dotnet test (LargeCapture · 4 GiB · Postgres)
-    runs-on: ubuntu-latest
+    # Pinned, not ubuntu-latest: on ubuntu-26.04 an unprivileged bwrap works, so the suites switch to confined mode; they assume the unconfined posture (confinement is covered by sandbox-isolation.yml)
+    runs-on: ubuntu-24.04
     # Three nested deadlines, ordered so the INNERMOST always fires first and the diagnosis is readable: the in-test
     # CancellationTokenSource at 40 m (AgentRunLogFourGibibyteCaptureFlowTests.cs) throws an XunitException naming the
     # stream and the SQL to inspect it; --blame-hang-timeout at 50 m only catches a genuinely wedged process, and