borg2: compact: optimize by caching used object ids of an archive

[ThomasWaldmann]

borg compact currently iterates over all archives, over all archived items, over all chunks in the items to flag the used chunks in the chunks index. Everything not flagged after that is not used.

Guess we could do that only once per archive and then cache the result: archive_id -> set of used chunks.

This could be very beneficial for people who have lots of large archives.

[ThomasWaldmann]

Guess HashTableNT could be used to efficiently represent these large sets.

[ThomasWaldmann]

Cache could live in the repository, so it is shared between clients.

[ThomasWaldmann]

I am working on this. #9829

Update: Closing this, needs too much memory.