After building packages, trigger CI to install and test

[ceball]

Currently, packages are built on travis/linux, then tested in various ways/environments, then uploaded to pyviz/label/dev and testpypi (for dev packages) and pyviz/label/main and pypi (for release packages).

Because we're using universal wheel/python:noarch, all platforms install the same packages. After building the packages on travis/linux, we should trigger travis/mac and appveyor/win to test the built packages (i.e. the packages users will be installing). (An alternative is travis/mac and appveyor/win building and testing the packages themselves - but then they wouldn't be testing the packages actually installed by users.)

Packages built on travis/linux could be stored on s3 and fully tested before ever uploading to pypi/anaconda.org, but it's probably simpler just to upload to pyviz/label/dev and testpypi, and learn that the packages are bad after that's happened. The same process could be used for releases, because a release should be "guaranteed to work" if it's just re-building the same commit that's already been through the dev packaging/testing (with a release tag rather than a dev tag this time).

This is separate from having appveyor/win and travis/mac test "develop install" - they can separately do that or not, depending what's best for a project.

[jbednar]

Sounds reasonable.