It's tricky to write unit tests for the CLI code for three main reasons:
exiting on errors, lack of dependency injection, and insufficient mocks.
argoproj#3917 tried to solve all
of these issues, but appears to have been abandoned. I thought about
trying to revive that PR, but it's been over 4 years and there's
certainly going to be a ton of conflicts. I figure it's better to take a
more focused, incremental approach.

This PR is focused on improving error handling in the following ways:
1. Do argument validation using [cobra
   validators](https://cobra.dev/#positional-and-custom-arguments)
   instead of custom logic in the `Run` function. Besides being
   simpler, this leads to better error messages.
2. Replace all `Run` functions with `RunE`, which allows returning
   errors. This is the same approach that Argo Rollouts is taking
   (example:
   https://github.com/argoproj/argo-rollouts/blob/00e39b114849010dd96221a8db441d58e860d4d0/pkg/kubectl-argo-rollouts/cmd/abort/abort.go#L41)
3. Replace nearly all calls to `log.Fatal()`/`errors.CheckError()`/etc
   with error propagation.

Signed-off-by: Mason Malone <mmalone@adobe.com>

Signed-off-by: Mason Malone <mmalone@adobe.com>

Signed-off-by: Mason Malone <mmalone@adobe.com>

Signed-off-by: Mason Malone <mmalone@adobe.com>

Signed-off-by: Mason Malone <mmalone@adobe.com>