Rather than walking the whole tree for each tag to find a possibly
corresponding row, use a hash table as cache.
This is a very significant improvement on large files with many tags,
reducing for example to about 170ms an update that took more than 18s
before.

Also fix merging of tags with same name and scope (probably unlikely to
exist in real-world files, but the tagmanager extract them correctly
and they used to display correctly too).


git-svn-id: https://geany.svn.sourceforge.net/svnroot/geany/trunk@5567 ea778897-0a13-0410-b9d1-a72fbfd435f5