entrypoint executes prematurely with --no-multicast-route

[rade]

$ weave launch-router
379393fd7be5641a635a58545c40163d07354825b9504f14fe7044950104d028
$ weave launch-proxy --no-multicast-route
8513c14984c5d52c8e712046dded770ceaff1aed72905004df07f4e14196af38
$ eval $(weave env)
$ docker run --rm -ti alpine true
Error response from daemon: Failure during network configuration for container e243ff04d0c2a73a6b9dac9785953a6607830a8ecfb0c159676c7ccd4791ee34:
nsenter: cannot open /hostproc/6873/ns/net: No such file or directory
$ docker run --rm -ti alpine ip addr show ethwe
477: ethwe: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1410 qdisc pfifo_fast state UNKNOWN qlen 1000
    link/ether 2e:87:cf:d1:0a:40 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::2c87:cfff:fed1:a40/64 scope link tentative
       valid_lft forever preferred_lft forever
Error response from daemon: Failure during network configuration for container 8263f808f03f863592f05f7650e4556c1a743436e0272a177571c19bcdf22328:
nsenter: cannot open /hostproc/8116/ns/net: No such file or directory

The problem is that we 'up' the interface way too early. This was introduced in 4311970. At the time --no-multicast-route didn't exist, thus masking the problem. Note that all call sites of connect_container_to_bridge up the interface at the correct time. And, afaict, none of the actions inside connect_container_to_bridge require the interface to be up. So the 'up'ing inside connect_container_to_bridge appears to be unnecessary and a mistake.