libblkid: care about unsafe chars in cache

libblkid: care about unsafe chars in cache

The high-level libblkid API uses /run/blkid/blkid.tab cache to
store probing results. The cache format is

   <device NAME="value" ...>devname</device>

and unfortunately the cache code does not escape quotation marks:

   # mkfs.ext4 -L 'AAA"BBB'

   # cat /run/blkid/blkid.tab
   ...
   <device ... LABEL="AAA"BBB" ...>/dev/sdb1</device>

such string is later incorrectly parsed and blkid(8) returns
nonsenses. And for use-cases like

   # eval $(blkid -o export /dev/sdb1)

it's also insecure.

Note that mount, udevd and blkid -p are based on low-level libblkid
API, it bypass the cache and directly read data from the devices.

The current udevd upstream does not depend on blkid(8) output at all,
it's directly linked with the library and all unsafe chars are encoded by
\x<hex> notation.

   # mkfs.ext4 -L 'X"`/tmp/foo` "' /dev/sdb1
   # udevadm info --export-db | grep LABEL
   ...
   E: ID_FS_LABEL=X__/tmp/foo___
   E: ID_FS_LABEL_ENC=X\x22\x60\x2ftmp\x2ffoo\x60\x20\x22

Signed-off-by: Karel Zak <kzak@redhat.com>