tailcat: fix home DERP race in the e2e tests

In the build sandbox there is no non-loopback interface, so magicsock
starts with its network down. The server's first netcheck, started by
SetPrivateKey, can run before tailcat calls SetNetworkUp(true) and then
never picks a home DERP. The server stays unreachable until the
periodic re-STUN 20-26s later, so clients time out after 10s. This made
TestServeExec and TestServeExitNode fail on x86_64-linux.

Mark the network up before SetPrivateKey.

Assisted-by: Claude Code (claude-opus-5-5)

pkgs/by-name/ta/tailcat/network-up-before-first-netcheck.patch

@@ -0,0 +1,29 @@
+Mark magicsock's network up before SetPrivateKey starts the first netcheck.
+
+With no non-loopback interface (e.g. the Nix build sandbox) the engine
+starts with magicsock's network down. SetPrivateKey and SetDERPMap kick
+off endpoint updates in the background, and if one of them runs before
+the later SetNetworkUp(true), updateNetInfo bails out on networkDown()
+without ever picking a home DERP. A server then is not reachable over
+DERP until the periodic re-STUN 20-26s later, and clients give up after
+10s with "tailcat Ping: context deadline exceeded".
+
+diff --git a/tailcat.go b/tailcat.go
+--- a/tailcat.go
++++ b/tailcat.go
+@@ -1577,6 +1577,7 @@ func (lb *locoBackend) Start() error {
+ 	mc := lb.sys.MagicSock.Get()
+ 	lb.logf("disco pub key: %v", mc.DiscoPublicKey())
+ 
++	mc.SetNetworkUp(true)
+ 	mc.SetPrivateKey(lb.priv)
+ 	mc.SetDERPMap(lb.dm)
+ 
+@@ -1632,7 +1633,6 @@ func (lb *locoBackend) Start() error {
+ 	mc.SetNetworkMap(nm.SelfNode, nm.Peers)
+ 	e.SetSelfNode(nm.SelfNode)
+ 	lb.sys.Netstack.Get().UpdateNetstackIPs(nm)
+-	mc.SetNetworkUp(true)
+ 	lb.logf("NetworkMap: %v", logger.AsJSON(nm))
+ 
+ 	// Install the live per-peer config sources. WireGuard peers are

pkgs/by-name/ta/tailcat/package.nix

@@ -19,6 +19,14 @@ buildGo127Module (finalAttrs: {
     hash = "sha256-Hue5NZAmAf8mxFqgoDcjt7YHiROeudpAoanxSFnZkTk=";
   };
 
+  patches = [
+    # The sandbox has no non-loopback interface, so magicsock starts with
+    # its network down. The server's first netcheck can then run before
+    # SetNetworkUp(true) and it never picks a home DERP, which makes the
+    # e2e tests time out on "tailcat Ping: context deadline exceeded".
+    ./network-up-before-first-netcheck.patch
+  ];
+
   vendorHash = "sha256-yfOl/gWIijLlqchXFiTRZ7vlgS/kn0xOmv52TFMYs+E=";
 
   subPackages = [ "cmd/tailcat" ];