lib: allow to specify registry and mediatype options

[lucab]

docker2aci (library and standalone binary), should allow the callers to specify a union-set of acceptable registry and mediatype options. This is needed to let consumers specify a certain mediatype to avoid (eg. OCI before stabilization) or certain registry features to enforce (eg. downgrade to registry-v1 or enforce DCT).

From the CLI, this may translate into:

• --registry-options=[registry-v1,registry-v2,registry-v2-dtc] with an auto default covering plain v1+v2 for the moment.
• --mediatype-options=[docker-v1,docker-v2.1,docker-v2.2,oci-v1.pre] with an auto default covering docker types for the moment.

For the lib, this will translate into appropriate parameters to all backend types.

[lucab]

I did an initial investigation into this, and it seems to require a non-trivial amount of reworking/cleanup both on the internal types and the exposed interface; for these reasons it was deferred post-0.13.

I'm assigning to @s-urbaniak for the design and implementation of the interface/types rework. CLI changes are nice to have but not required at first.

As we are most probably changing the external lib interface, it would be nice to already start to account for:

• passing a logger interface to the lib: Allow custom logging in docker2aci lib. #136
• different streams for warnings and diagnostic messages: Provide warnings/errors during conversions #24
• stream targets set by the caller: Let user specify the stream to print messages #61
• streaming the result of conversion directly, without writing to a file: Add ability to stream output #93

[cgonyeo]

I don't like the concept of these flags, docker2aci should be able to auto-determine (and currently can) which mediatypes to use in a given situation. If v2 pulling is supported, it'll do v2 pulling accepting both v2.1 and v2.2 content, and if v2 pulling is unsupported it will automatically fall back to V1 pulling. I'm not sure a user would ever want to selectively disable one of these.

I'm currently making a PR for hiding OCI support behind a flag, but I've got other thoughts on that I'll describe in that PR.

[s-urbaniak]

Cross-referencing #236 (comment) as it describes the rationale behind this issue.