test(e2e): unmanage the TAP anchors in the remaining kernel scenarios

Same guard the ethernet-switch scenario already carries: NetworkManager
can release an NM-managed TAP from its bridge shortly after
(re-)enslavement (docs/bugs/networkmanager-tap-release.md). Applied to
the Dynamips, QEMU, IOU, IOL-runner and switch-cascade scenarios after
their nodes have booted (no-op where nmcli/NetworkManager is absent).
This keeps those tests deterministic on un-declared hosts.

tests/e2e/test_dynamips_kernel_datapath.py

@@ -111,6 +111,11 @@ def test_dynamips_kernel_datapath():
         c1 = _boot(server, pid, r1_id)
         c2 = _boot(server, pid, r2_id)
 
+        # NetworkManager can release an NM-managed TAP from its bridge after
+        # (re-)enslavement (docs/bugs/networkmanager-tap-release.md); take the
+        # anchors out of its reach for the rest of the run.
+        harness.unmanage_from_networkmanager(t1, t2)
+
         # Kernel objects: anchors exist and the per-link bridge has exactly
         # the two of them enslaved.
         assert harness.tap_exists(t1) and harness.tap_exists(t2), (t1, t2)

tests/e2e/test_ethernet_switch_cascade.py

@@ -188,6 +188,11 @@ def test_ethernet_switch_cascade_kernel_datapath():
         c1 = _boot(server, pid, r1_id)
         c2 = _boot(server, pid, r2_id)
 
+        # NetworkManager can release an NM-managed TAP from its bridge after
+        # (re-)enslavement (docs/bugs/networkmanager-tap-release.md); take the
+        # anchors out of its reach for the rest of the run.
+        harness.unmanage_from_networkmanager(a1_10, a1_20, a2_10, a2_20)
+
         # The deferred host joins completed via the node-start re-push: each
         # switch bridges its two router anchors plus its cascade end.
         assert harness.wait_until(lambda: harness.bridge_members(br1) == sorted([a1_10, a1_20, e0]), timeout=15), (

tests/e2e/test_iol_docker_kernel_datapath.py

@@ -121,6 +121,11 @@ def test_iol_docker_kernel_datapath():
         c1 = _boot(server, pid, n1_id)
         c2 = _boot(server, pid, n2_id)
 
+        # NetworkManager can release an NM-managed TAP from its bridge after
+        # (re-)enslavement (docs/bugs/networkmanager-tap-release.md); take the
+        # anchors out of its reach for the rest of the run.
+        harness.unmanage_from_networkmanager(a1, a2)
+
         # Kernel objects: anchors exist (born with the node, before the link
         # was ever wired) and the per-link bridge has exactly the two of them.
         assert harness.tap_exists(a1) and harness.tap_exists(a2), (a1, a2)

tests/e2e/test_iou_kernel_datapath.py

@@ -135,6 +135,11 @@ def test_iou_kernel_datapath():
         c1 = _boot(server, pid, n1_id)
         c2 = _boot(server, pid, n2_id)
 
+        # NetworkManager can release an NM-managed TAP from its bridge after
+        # (re-)enslavement (docs/bugs/networkmanager-tap-release.md); take the
+        # anchors out of its reach for the rest of the run.
+        harness.unmanage_from_networkmanager(a1, a2)
+
         # Kernel objects: the anchors are one persistent TAP per Ethernet
         # bay/unit, born with the node (all eight — the IOU port model),
         # and the per-link bridge has exactly the two addressed ports.

tests/e2e/test_qemu_kernel_datapath.py

@@ -127,6 +127,12 @@ def test_qemu_kernel_datapath():
         c1 = _boot(server, pid, n1_id)
         c2 = _boot(server, pid, n2_id)
 
+        # NetworkManager can release an NM-managed TAP from its bridge after
+        # (re-)enslavement (docs/bugs/networkmanager-tap-release.md); take the
+        # anchors out of its reach for the rest of the run (the restart below
+        # re-attaches them under fresh devices).
+        harness.unmanage_from_networkmanager(a1, a2, harness.qemu_anchor_name(n1_id, 1))
+
         # Kernel objects: one persistent TAP per adapter (both adapters of
         # router 1, the addressed one of router 2), and the per-link bridge
         # has exactly the two addressed ports.