a bit more normalization

[eroux]

I think in order to convert archaic to modern you also have to spot some patterns with regexps (because the list would be too long)

• da drag, it could be either detected by regexp or just taking a list like this one (the da drag are not written, but they are attested in all these words)
• suffix འ could also be removed when not in the small set of syllables it can be attached to (see this article, section 2.2.7)
• Other patterns have been isolated by a project on Old Tibetan, recorded here and implemented also here / here