Rework collation generation:

When building collation database for non unicode encodings use the proper
unicode mapping (this fixes collation not working properly for those encodings)

For locales where new characters are added but only for unicode, stop trying to
map the new characters, directly extract from CLDR the collation files for the
said encoding

Stop trying to generate encoding map from unicode version for GB2312 and encCN
It was not reliable. Instead use the map provide by the CLDR project

Reported by: ache