Going forward, there will only be two input methods, one taking raw data
and one taking a variant with already decoded data. This is also continuing
the work of phasing out ExtractorInput enums in favor of full MIME types.

Nothing really changed on the inside yet, this is mostly transitional
scaffolding.