weave start with container name or short id leaks allocated IPs and DNS records

[rade]

To reproduce

$ weave launch -iprange 10.3.0.0/24
c86dea6734a9bee5788e07a9cafa064178d5f930d4582e60d8d59ef7f7f3604b
$ weave launch-dns 10.3.1.1/24
6615e65710c480b30f492b03529b70313e7b6a89e31719f3274ab08d2b1e5328
$ docker create --name=foo -h foo.weave.local -ti ubuntu
ba6622c1301f614d5e38422324a718199629f45b6e22331b163e2878edd71b71
$ weave status | grep -E '(Free IPs|foo.weave.local)'
Allocator subnet 10.3.0.0/24  Free IPs: ~99.2%, 254 local, ~0 remote
$ weave start foo
foo
$ weave status | grep -E '(Free IPs|foo.weave.local)'
Allocator subnet 10.3.0.0/24  Free IPs: ~98.8%, 253 local, ~0 remote
foo: foo.weave.local.[10.3.0.1]
$ docker kill foo
foo
$ weave status | grep -E '(Free IPs|foo.weave.local)'
Allocator subnet 10.3.0.0/24  Free IPs: ~98.8%, 253 local, ~0 remote
foo: foo.weave.local.[10.3.0.1]

The problem stems from the fact that docker start just echoes back what it is given, rather than the full container id. Which is most useless.

This was spotted by @bboreham

[rade]

The fix for this is trivial but we really should add some tests, at least for DNS (which should be easy; just check for a failing lookup). IPAM is harder.

[bboreham]

On the multi-subnet IPAM branch (#827), the fix is 064dfc6, the DNS test is c4f86fc and the IPAM test is part of eb92275