A system_clock::time_point can't be used to represent all valid values
of a 64-bit time_t because nearly half the bits are used up for the
nanoseconds part. This makes filesystem::last_write_time() check the
time_t values and report an error if it can't be converted to the
time_point type. This means we don't get undefined behaviour for files
with mtime after the 24th century :-)

Thanks to Eric Fiselier for point it out.

	* include/experimental/bits/fs_fwd.h (file_time_type): Simplify
	definition.
	* src/filesystem/ops.cc (file_time): Take error_code parameter and
	check for overflow.
	(do_copy_file, last_write_time): Pass error_code in file_time calls.
	* testsuite/experimental/filesystem/operations/last_write_time.cc:
	New.
	* testsuite/util/testsuite_fs.h (scoped_file): Define RAII helper.

Tested x86_64-linux, committed to trunk. Backports for this (and some
other filesystem changes) to follow soon.