Do two passes instead of just sequentially writing all chunks to the standard output.
On the first pass, all "chunk emit" instructions are remembered together with their output
positions indexed by bundle id, and all "byte emit" instructions are executed using seeks.
On the second pass, all remembered "chunk emit" instructions are executed in the bundle
order. This makes zbackup decompress every used bundle only ONCE instead of doing
it (basically the same work) many times while reading different chunks.
This allows for bigger bundle sizes (I use 32M), which reduces the number of files
in the repository and makes it more cloud-storage-sync friendly, and further improves
the compression ratio.
Re-uses most of the Mac OS X changes, but adds a read(2)/write(2) loop instead
of sendfile(2) when renaming across devices, because OpenBSD doesn't have
sendfile(2).
Vitaliy Filippov
Ulrich Alt