[WIP] Add 'skip_port_validation' flag

[stephenfin]

This allows us to skip port validation for images that publish multiple
ports, not all of which may be active at any given time. We also
introduce reno, which is a release note manager that we can use to
produce a helpful changelog for users.

This is marked WIP since I haven't tested it and it needs unit tests.

[dcrosta]

This looks reasonable, though it is a big hammer that will probably cause a lot of frustration. When I introduced the health check support in #30, I intentionally left the port checking in, even for containers that have a health check. I'm not sure what's the right call, so I opted for the "better safe than sorry" route.

Would it also have worked for your use case (I assume this is in response to #35?) if you could specify specific ports to publish, similar to docker run -p X:Y, and then only port-test that list?

[stephenfin]

This looks reasonable, though it is a big hammer that will probably cause a lot of frustration. When I introduced the health check support in #30, I intentionally left the port checking in, even for containers that have a health check. I'm not sure what's the right call, so I opted for the "better safe than sorry" route.

Definitely a big hammer and I'm not sure it's the correct approach (which is why I didn't invest too much time on writing tests yet).

Would it also have worked for your use case (I assume this is in response to #35?) if you could specify specific ports to publish, similar to docker run -p X:Y, and then only port-test that list?

The ability to specify the specific ports that should be published would work ideally actually and I think I'd rather it to this approach. Would we be providing a container-host mapping (X:Y) or could we let the host-side of things be auto-negotiated as happens when you publish all ports (so you'd continue inspecting the environment variables)?

[dcrosta]

We should let docker pick an appropriate port, and discover it the way we do today (from container.attrs). The [docker:whatever] section should get a new list called publish_ports or something like that listing a subset of the EXPOSEd ports.

It probably should be an error if you ask to publish a port the container doesn’t expose, if that’s possible.

[dcrosta]

This is superseded by #49, just merged.