new subcommand: opam admin migrate-extrafiles

[hannesm]

This subcommand allows to move all extra-files of an existing opam repository into extra-sources. The files are extracted into a specified local directory, and the opam files are edited in-place.

see #5811 for motivation

[rjbou]

Thanks for the PR! We'll look at it once opam 2.2 release is done

[hannesm]

For the future reviewer, a slightly modified piece (see hannesm/opam@migrate...hannesm:opam:migrate-extra-files) was used for driving the rewrite for opam-repository (ocaml/opam-repository#25960)

The main differences are:

• if SHA512 is used as checksum, keep it (no need to replace with SHA256)
• for easier reviewability, keep existing MD5 checksums (I even wrote another tool, https://github.com/hannesm/opam-check-checksum, to check that) of extra-files, and add the SHA256 to extra-source next to the MD5
• file naming: after more thoughts: deduplcation is great, but for a future developer, it is great if there are e.g. 25 META files for a single package, they should have reasonable names -- so the migration uses the META.PKG_VERSION (and at the point when it is not able to deduplicate everything, it just duplicates the same file if necessary) -- to support this, a two-stage approach was used, and the "not-to-be-deduplciated-files" are part of the commits shown in the link above.

So, it boils down to what to expect from such an automated migration, and whether there's need to have the result conveniently editable by people (or by scripts). Since some choices are hardcoded (such as the hash algorithm -- the tool has only ever been applied with SHA256 (and there are special code paths for "what if an MD5 is present" and "what to do if a SHA512 is present"), which is unlikely something you'd want to embed into the opam source tree.

About file naming, for deduplication reasons (and please note, there are other ways to achieve deduplication), the layout on the opam-source-archives repository is "patches/pkg-name/file1..N", i.e. not another subdirectory for pkg_version.

[kit-ty-kate]

I've rebased and made some improvements to the UI of the command but i haven't reviewed the core of the code yet, although it looks fine on first read.