Replacing analysis code with uralicNLP

[nikopartanen]

Ideally each speaker's tier would be sent to analyser in one go.

[nikopartanen]

@meehkal and @jeutzsch, you probably have some thoughts on this. I recently implemented this change within the file:

https://github.com/langdoc/elan-fst/blob/master/elan_fst.py

And it seems to work very well, there are following advantages:

• We can run multiple FST's and then put that through one CG, i.e. analysis on kpv + koi and CG with kpv.
• This also gives us the language tags, if those would want to be inserted
• Local installations are not needed
• Does not use xfst anywhere
• We also get syntactic readings now (for Komi there are so few that inserting them is not maybe necessary)

Disadvantages:

• Can be used for languages not supported by uralicNLP, but this is not very well documented, and involves putting compiled transducers to the locations where uralicNLP finds them. In principle for old setup any compiled FST could had been given.

Things to do:

• Connection to the lexical resources is more complicated question, but needs to be addressed if we want to get translations of the lemmas to the ELAN file. This is also really wide topic where lots of people are already doing much work, so instead of active doing one should sit and wait that things just progress to the point they can be integrated here.