Until the GVFS.Service process is implemented on Linux,
the "WARNING: ..." message printed when it cannot be contacted
at mount time has the side-effect of confusing the some
functional tests, as the output from the GVFS process is scanned
for any string matching "warning" case-insensitively.

To avoid this, we force the GVFS_UNATTENDED environment variable
to be set on Linux for all functional tests, which in turn
causes the GVFS process to skip checking for a running GVFS.Service
process.

Stub out the call to PlaceHolderHasVersionInfo() on Linux,
akin to what is currently done for the Mac.

Also enable the same code on Linux as is used on Mac for
the EnumerateAndReadDoesNotChangeEnumerationOrder()
functional test.

These three tests currently fail due to contention with
libprojfs for an exclusive flock(2); the C# file locking
implementation on Linux conflicts with the locks taken by
libprojfs before testing the projection status of a file.

We can remove this TODO category once libprojfs uses an
internal locking mechanism instead of flock(2).

Where appropriate, rename the categories of functional tests
which have known failures on both Mac and Linux as
NonWindowsTODO instead.

Several tests currently fail due to apparently delayed file
writes on Linux when using the FUSE passthrough filesystem
implemented in libprojfs.

We should remove this TODO category once libprojfs and FUSE
file write buffering is investigated and remediated.

On Linux, when using libprojfs (which for the time being
internally calls flock(2) to acquire its per-inode
virtualization state), EAGAIN/EWOULDBLOCK may be returned
while writing to files in MultithreadedReadWriteTests.

On Linux, the TruncatedObjectRedownloaded() functional test
requires that we trigger hydration by actually reading the
contents of the file whose GVFS cached object file we've
artifically truncated, rather that just checking that the
file exists, which is sufficient on macOS for the test to
succeed (but not sufficient on Linux).