reconsider app init when testing CLI commands

[rsyring]

Currently, testing.invoke_command() sets the environment variable USE_TEST_PROFILE which triggers the use of, wait for it, the testing profile. But, .cli.CLILoader.create_app() inits the app directly and doesn't use Keg.test_prep() which means the app instance is not cached and the testing_run_start signal might not get sent.

It seems like a refactoring of Keg.init() to support the testing_prep() logic might be called for and that logic could be triggered by a keyword arg or environment variable. One caveat though is that when using invoke_command() args might be sent to the app that affect it's instantiation. If they are different, you wouldn't want to return the cached app.