Add ability to "pause" a pod, so that possibly crash-looping containers can be inspected manually

**What would you like to be added**:

The ability to "pause" a pod, so that possibly crash-looping containers can be
inspected manually.

**Why is this needed**:

There are multiple reasons for containers to crash-loop. A common one is: the
specified container command crashes, causing the container to fail and be
restarted. In situations of infinite crash-loops under some back-off strategy,
it is very difficult to investigate the state of a container:
- its files
- environment variables
- command outputs

Being able to start a container and not run its command, but do something like
sleep instead would greatly help with debugging lots of issues.

There might be multiple ways of implementing this. A hack I've been using as an
operator developer is just commenting a containers original `command` field
and replacing it with something like:

``` yaml
command:
- bash
- -c
- "while true; do echo 'sleeping...'; sleep 5; done"
```

Start with the KEP process linked in the issue and investigate how Kubernetes pods and container commands can support a paused debugging state. Define what "done" means as starting a crash-looping container without its original command while allowing inspection of its files, environment variables, and command output.

Written by the indexing model from the issue text.