The snapshot id doesn't make sense when two snapshots are squashed.
Better to use the timeline id anyway.
It's still a little strange to squash two timelines together, but it
must be possible to flatten history even if there have been many branch
events, so this is should be a control plane / snapshot management
problem.
Add logic to squash snapshot files.
Add snaptool (a binary for inspecting and manipulating snapshot files).
Use bookfile 0.3, which allows concurrent reads.
The snapfile crate implements a snapshot file format. The format relies
heavily on the bookfile crate for the structured file format, and the
aversion crate for versioned data structures.
The overall structure of the file looks like this:
- first 4KB: bookfile header
- next 8KB * N: raw page data for N pages
- page index map (from page identifier to data offset)
- bookfile chapter index
When a SnapFile is opened for reading, the page index map is read into
memory; any page can be read directly from the file from that point.
