Conflicts:
	ft/ft-ops.cc

FT-633 test case and bug fix for the data race between the lock request start and retry functions that results in a stall in the condition var broadcast

leftmost le (oops!). Also strengthen our unit test for coverage.

DRD seems to prefer those annotations to
DRD_IGNORE_VAR/DRD_STOP_IGNORING_VAR.

Helgrind is stupid, and VALGRIND_HG_{DIS,EN}ABLE_CHECKING pairs
bizarrely "reclaim" the memory location for the current thread when
re-enabling checking.  So they probably aren't completely suitable for
our purposes, but Helgrind provides no facility for doing what we want
because it just wants to watch the world burn.  So we add a couple of
Helgrind suppressions in addition.

Also, added a cute RAII wrapper for reading a variable in a racy way.
Writing can be done the same way if we only want to appease DRD, but
because Helgrind does things on a memory location-basis and not a
thread-basis, things get a little tricky.

either use all-plain or all-keyword style of target_link_libraries

also widened one valgrind suppression to also catch something we
initialize in the assert layer

Newer libc doesn't like _SVID_SOURCE:
https://sourceware.org/ml/libc-alpha/2013-12/msg00652.html

variables with `drd ignore var' and `drd stop ignoring var'

FT-594 add fractal tree version 24 (tokudb 7.0.1) and version 27 (tokudb 7.5.0) stress data directories

bits were inconsistent though the handle was opened without those bits
set. This caused dbremove to always fail.

FT-591 fix valgrind uninitialized value error in block allocator test caused by reading past the end of a the blockpair array

FT-273 Prevent setting the memcmp magic on db handles for which a FT is already open. Improve comments. Add a test.