Namespace forking

[MarinPostma]

This PR introduces a basic forking functionality. The admin API route POST /v1/namespaces/:from/fork/:to forks the namespace :from into a new namespace :to from the current point in time.

Forking works by creating a copy of the source database. This copy is performed from the replication log, to ensure point-in-time consistency. The forking algorithm works as follows:

• Determine the point in time at which to fork. This is done by saving the current frame_no.
• Replay the replication log from the start until the chosen point, and write that to a new database file
• Create a new namespace by placing the database file in the correct directory
• Let the recovery process take care of regenerating a replication log and replicating to bottomless

[psarna]

Note: for a cheap copy, we could use XFS' (and some other fs') capabilities of copy-on-write, based on https://man7.org/linux/man-pages/man2/ioctl_ficlone.2.html . Probably not trivial with our current design where a restore also takes place, it's just an interesting concept - e.g. perhaps we could CoW-copy the wallog.

[MarinPostma]

@psarna we will iterate on that design for sure

[haaawk]

Is anything blocking this PR? @psarna @MarinPostma ?

[MarinPostma]

@haaawk nope, just waiting for review

[psarna]

what does this bool mean? Please add a comment

[psarna]

Ok, now I went over the whole change and I'm even more confused - why was this parameter added at all? It looks like it's not used anywhere, you just ignore it above.

[MarinPostma]

it is used there, I will add a docstring
https://github.com/libsql/sqld/pull/646/files/3b76a120da75aaf2794f4f50f039a4b35b4e6c8d#diff-3035393d5fdf2fb87106000360e07ea7386f5ef923d280dd6eea3dfc9181ba39R453