Two patches:

1. As it is now, mpg123 backend reports -1 duration for multiple mp3s
for me.  The failure is because we skipping the tags unconditionally
since changeset 1072:503ad6c93c50.  If libmpg123 sees the ID3v2 tags,
it does properly report duration for those files.  So the first patch
lets music_mp3123 to keep the ID3v2 tags.

2. Changes to mp3_skiptags() so that it consumes all the tags at file
end:  We do not know the order of ape, or lyrics3, or musicmatch tags,
so we loop until we consume them all, scanning for each tag type once.
(I don't yet care about freaky broken mp3 files with double tags.) It
also removes inline directive from a few detection procedures there.

OK to apply?