Get rid of parsers-megaparsec

[willbasky]

It not maintained and keeps dependencies outdated.

[willbasky]

radixtree depend on Parsec. Either get rid of radixtree (swap it with another one) or fork and base radixtree on megaparsec or fork (or make PR) and update parsers-megaparsec

[willbasky]

To complete the task it wanted to swap Radixtree's search to lookup because search requires constraint from parsers-megaparsec while lookup doesn't.

search gives parsing of dirty strings, e.g.

> parseWylieInput radix "(balka)"
> parseWylieInput radix "(balkana)"
ExceptT (Identity (Right [([],[["balka"]])]))

it reads radixed words and drops non-radixed in anyway. Therefore dirty wylie text should parsed better beforehand. If it parsing of dirty wylie text not needed anymore, then 'lookup` will be enough. There are two approaches

1. Add non-wylie chars and strings to sillabies and then parse (1.balka) -> ["(", "1", "." "balka", ")"] and then map lookup it and it becomes "(1.བལྐ)" or "(༡.བལྐ)", etc...
2. Parse dirty text by separating non-wylie stuff from wylie (1.balka) -> ["(", "1", "." "balka", ")"] and then map lookupLemient (lookupLemient left non-wylie untouched) and convert wylie to tibetan script ("(1.བལྐ)" or "(༡.བལྐ)", etc...)

   
   lookupLenient :: RadixTree a -> Text -> Text
   lookupLenient radix t = case lookup radix t of
   Nothing -> t
   Just (t',_) -> t'

map (lookupLenient radix) ["(", "balka", ")"] ["(","balka",")"]

In both 1 and 2 a dirty text must be separated roughly,
hence first approach has no advantages,
and it is better to keep syllabies non-wylie-free
therefore second approach is right choice.

Steps of realization:

1. Reduce current transforming to flat list. Some texts won't be consumed, and be dropped.
2. Refactor current parsers to more correct separation against flat list.