In the past we've got situations when the CI integration tests reports
were not extensive enough to clarify what's going on, and we were
launching a full-blown debugging process just to find out that the
probes were missing.

To make initialization issues more visible, make integration tests
verbose about them:

* Introduce a HEALTHCHECK for the Collector docker image, based on the
/ready Civet API. This command would be ignored on K8S, but we could use
it on CI for containers filtering.

* Wait for Collector container to become healthy first before proceeding
with the rest of tests. This will make it immediately clear if the test
is failing due to missing probes.

* Make self-checks verification a hard failure. Currently, we wait for
the self-checks, but do not report if they're missing. This change will
make it clear if something is badly broken to the extent that we don't
receive any events.

Taking into account volatile nature of integration tests, it's
improbable that exactly N processes could be observed during any test.
At the same time we actually want to assert that at least N processes
were observed (e.g. at least 1 to make sure we've got self-checks).
Modify the condition to follow "at least N" approach.

Container doesn't have xargs.

Reduce waiting threshold, make errors more verbose.