But generate Warning events identifying the missing Services.

Revert "Revert "Revert "Revert "Revert "Revert "Revert "Comment-only change to trigger the gitlab CI test stage."""""""

This reverts commit 8f759271.

XXX: consider adding another static analysis tool.

Revert "Revert "Revert "Revert "Revert "Revert "Comment-only change to trigger the gitlab CI test stage.""""""

This reverts commit c107da04.

Revert "Revert "Revert "Revert "Revert "Comment-only change to trigger the gitlab CI test stage."""""

This reverts commit e600593d.

This reverts commit 77f1dfd3.

This reverts commit 4117718e.

This reverts commit c1fca90b.

This reverts commit 666e3373.

This reverts commit 5b0381e5.

For this we make the event generating functions public, and move
their definitions into varnish.go, where most of the Controller
class is defined.

- k8s client-go v0.25.5
- Build with go 1.19

While here:
- prometheus client v1.15.1
- logrus v1.9.0

For the client-go update, two changes were necessary:

- The workqueue MetricsProvider must satisfy a new interface.
- The getter method for Pods requires a context argument.

Other updates:

- go version 1.15
- Varnish 6.6.2
- debian buster base image

Previously only files specified in the ConfigMap were checked for
changed modification times. If only generated VCL sources were
changed, usually due to Endpoint changes, and the VCL load failed
for any reason, tehn afterward no changed VCL sources or changes
in the ConfigMap were detected. This could lead to VCL never being
updated after a change only in the generated VCLs, although the
config does not correspond to the desired state of the cluster.

Now we keep a record of the latest mtime of all VCL sources,
including generated sources, and reload VCL if any of them are
currently newer.

Types for literal structs don't need to be named when they are
unambiguous.

At info level, log the latest known VCL mtime. Log all of the mtimes
at debug level.