can iterate over and clean up, which prevents a memory leak when the
fractal tree shared library unloads before all thread-local destructors run

This reverts commit 843f0050.

Should have been on a branch.

This reverts commit 7fa771e0.

Should have been on a branch.

This reverts commit fbd96859.

Should have been on a branch.

This reverts commit d467d3cf.

Should have been on a branch.

the last point we may read the node variable safely.

TOKU_GLOBAL_STATUS

disk bytes written. Need to add a 'double' engine status type, which
apparnetly exists in the union, but not as a type tag.

node and notices that it is over 32mb

This reverts commit b1c1e93d, reversing
changes made to 60be90af.

Instead just keep mempool around a little longer (like pre-refactoring).  This should
reduce large mallocs and remove a memcpy

Refs Tokutek/ft-index#46 Comments, minor renames, reworked bn_data::move_leafentries s.t. splits do not significantly increase memory usage of basement nodes and to be cleaner

now uses dmt iterate_on_range and the dmt builder instead of a for loop

Refs Tokutek/ft-index#46 Killed dmt_functor<> template.  Added dmtwriter_t template parameter to dmt and some documentation

Refs Tokutek/ft-index#46 Add some comments.  Improve dmt::verify() to check for more types of corruption.  More static checks

Refs Tokutek/ft-index#46 Remove ctree/dtree/mnode/cnode/base node/dnode/marks/etc... That is, removed all items that
were designed to allow us to upgrade the dmt later (e.g. fixed-length nodes for tree form) or merge with old omt (e.g. supporting marks)

Use dmt to replace omt in bn_data class for storing leafentries.
Optimization for serial inserts and mempool

osx doesn't support __thread well so scoped_malloc breaks it.  Rather than
fix it, we implement a correct but poorly-performing implementation for
now that will be used on osx machines.

fixes #160

Refs Tokutek/ft-index#159 Calculate klpair size correctly when message application deletes a leafentry