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.

Should solve the conflict between read-only checks and housekeeping which
caused test failures.

This one has the "Potential" to be come the permanent database, by calling
Commit() on it. Soon there will be real "temporary" databases which live in the
temp directory and cannot be committed.

Attempt to fix 32-bit and 64-bit cmake builds