Refactor the publish.Namer, add --bare option for image naming

[mattmoor]

In some environments (e.g. ECR) each repository needs to be pre-created, which gets tedious.

Given that ko does everything by digest, this collapses all of the images published into the same repository: ${KO_DOCKER_REPO} with no suffix to discriminate between images.

This is also beneficial from a security standpoint in certain contexts, as KO_DOCKER_REPO=docker.io/{user} might publish new repositories as public by default.

[mattmoor]

For example with ECR, my registry domain is 912237561242.dkr.ecr.us-west-2.amazonaws.com, and I have to create each of: 912237561242.dkr.ecr.us-west-2.amazonaws.com/{foo} today.

With this, I can use KO_DOCKER_REPO=912237561242.dkr.ecr.us-west-2.amazonaws.com/ko-images and ko apply --naked will publish all images to the above repository and substitute their digests into the yamls, e.g. 912237561242.dkr.ecr.us-west-2.amazonaws.com/ko-images@sha256:deadbeef

In these cases, no part of the import path is incorporated into the repository name, which is where the name "naked" comes from. I'm open to alternative naming suggestions, but struggled to find a more descriptive term than this.

[jonjohnsonjr]

My only real issue with this is having a global flag called --naked. I wish we had a separate mechanism for selecting a namer that wasn't flag-based, ref:

#172
#143

[mattmoor]

@imjasonh also doesn't like the name either. He suggested --bare, which I think bears the same meaning as --naked, but without the same (level of) vulgar connotation 🤷

I am happy to switch to --bare, though I'd probably drop the short flag since we already have -B.

[jonjohnsonjr]

I'd prefer dropping the short flag regardless, since they're such a precious resource. In fact, it would be great if this was really cumbersome to type so that you got annoyed into implementing a better mechanism for this :)

Some dumb ideas that might inspire better ideas:

• --flat
• --direct
• --nosuffix
• --norepo[sitory]
• --noname
• --nemo

In this mode, we should probably warn or error if you pass a single tag and we publish multiple images, because the tag will get overwritten. Alternatively (lazily), just document this caveat in the README or flag description.