Should hopefully reduce risk of random test failures/race conditions.

Adding individual -DBOX_CURRENT_FILEs to each file's compilation flags worked, but caused
each command to be different, massively increasing verbosity of CMake builds. Since there wasn't
apparently a compile-time solution that works with VC++, I've gone back to the slower runtime way,
but only computing the value when needed (when per-module logging overrides are in use).

Should hopefully reduce random failures of timing-dependent tests which
rely on subsecond precision for accuracy.

Takes too long in release mode, due to the S3 block size.

This avoids abusing "make install" to install them in the correct locations
for running tests, which causes problems for CPack installers which rely on
the list of install()ed files.

Otherwise "make install" fails during tests because it can't write to /usr/local.

For some reason it's becoming increasingly common for Unixes not to set TMP* to
anything.

flock and fcntl(F_SETLK) locks are not exclusive within the same process,
so this test cannot work as designed platforms that don't support a more
robust form of locking such as O_EXLOCK or Windows locks.

Fix test/common for NamedLock by starting a separate process so that we can
really test conflicting locks.

The abbreviated name doesn't work for 32-bit compiles!

Use PLATFORM the way that MSBuild wants it to be used, and add Win64 to the
CMake Generator name using PowerShell instead.

Hopefully will enable Win64 builds. At least it works for me on a VM.