Remove vn_rename and vn_remove dependency

The only place vn_rename and vn_remove are used is when writing
out an updated pool configuration file.  By truncating the file
instead of renaming and removing it we can avoid having to implement
these interfaces entirely.  Functionally an empty cache file is
treated the same as a missing cache file.  This is particularly
advantageous because the Linux kernel has never provided a way
to reliably implement vn_rename and vn_remove.

The cachefile_004_pos.ksh test case was updated to understand
that an empty cache file is the same as a missing one.

The zfs-import-* systemd service files were not updated to use
ConditionFileNotEmpty in place of ConditionPathExists.  This
means that after exporting all pools and rebooting new pools
will not the scanned for on the next boot.  This small change
should not impact normal usage since pools are not exported
as part of a normal shutdown.

Documentation was updated accordingly.

Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Arkadiusz Bubała <arkadiusz.bubala@open-e.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes openzfs/spl#648 
Closes openzfs#6753

module/zfs/spa_config.c

@@ -181,12 +181,10 @@ spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 	 * If the nvlist is empty (NULL), then remove the old cachefile.
 	 */
 	if (nvl == NULL) {
-		err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
-		/*
-		 * Don't report an error when the cache file is already removed
-		 */
+		err = spa_config_remove(dp);
 		if (err == ENOENT)
 			err = 0;
+
 		return (err);
 	}