instead refer to tokuft

one large buffer to serialize a node instead of many smaller ones, which
should hopefully put less pressure on jemalloc during checkpoints etc.

cachefiles, also clean up some formatting.

up report formatting a bit. Also strengthen malformed trace detection.

a running trace (that is, a trace that is still getting written to by some
process)

fragmentation changed

tracing code

ba_replay

create_from_blockpairs would crash the replay. Also fix parsing bugs.

multiple calls), support create_from_blockpairs in the allocator and
trace tool

test

--get 0 test failure

Also, tidy up some indentation that got messed up during the refactor.

the osx build)

initialization code

likely the allocation will need to move again at the next checkpoint (we
pass the node height for this value). The new heat zone allocation
strategy uses the heat value to put nonleaf nodes towards the end of the
file and leaf nodes towards the beginning.

- Failure to handle multiple allocators at once (oops)
- Failure to handle a clean shutdown where allocators are destroyed by
  the trace gracefully and so we can't get stats for them after the run
  (need to clean this up eventually)
- Added line numbers to error messages so debugging is easier

allocator trace gets written to