net/udprelay: change Server.AllocateEndpoint existing alloc strategy (#15792)

The previous strategy assumed clients maintained adequate state to
understand the relationship between endpoint allocation and the server
it was allocated on.

magicsock will not have awareness of the server's disco key
pre-allocation, it only understands peerAPI address at this point. The
second client to allocate on the same server could trigger
re-allocation, breaking a functional relay server endpoint.

If magicsock needs to force reallocation we can add opt-in behaviors
for this later.

Updates tailscale/corp#27502

Signed-off-by: Jordan Whited <jordan@tailscale.com>

net/udprelay/server_test.go

@@ -174,8 +174,7 @@ func TestServer(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	// We expect the same endpoint details as the 3-way bind handshake has not
-	// yet been completed for both relay client parties.
+	// We expect the same endpoint details pre-handshake.
 	if diff := cmp.Diff(dupEndpoint, endpoint, cmpopts.EquateComparable(netip.AddrPort{}, key.DiscoPublic{})); diff != "" {
 		t.Fatalf("wrong dupEndpoint (-got +want)\n%s", diff)
 	}
@@ -191,6 +190,15 @@ func TestServer(t *testing.T) {
 	tcA.handshake(t)
 	tcB.handshake(t)
 
+	dupEndpoint, err = server.AllocateEndpoint(discoA.Public(), discoB.Public())
+	if err != nil {
+		t.Fatal(err)
+	}
+	// We expect the same endpoint details post-handshake.
+	if diff := cmp.Diff(dupEndpoint, endpoint, cmpopts.EquateComparable(netip.AddrPort{}, key.DiscoPublic{})); diff != "" {
+		t.Fatalf("wrong dupEndpoint (-got +want)\n%s", diff)
+	}
+
 	txToB := []byte{1, 2, 3}
 	tcA.writeDataPkt(t, txToB)
 	rxFromA := tcB.readDataPkt(t)