Move parentheses usage logic from AST to printer

[ark120202]

• Having specific nodes for parentheses doesn't make much sense for me. Parentheses are supposed to be a tip to parser, in AST we don't have any ambiguities where parentheses matter:

   (1 + 2) + 3
  
   BinaryExpression:
   	left:
   		BinaryExpression:
   			left: 1
   			right: 2
   	right: 3
  

   1 + (2 + 3)
  
   BinaryExpression:
   	left: 1
   	right:
   		BinaryExpression:
   			left: 2
   			right: 3
  

• It would make code cleaner and easier to maintain, since to fix issues like Fix for property access on string and array literals #650 we won't have to duplicate logic everywhere
• It would make code less bug-prone, since we'll have less invalid states in AST
• Most of JS parsers (babel, acorn, esprima) do it

[ark120202]

Resolved in #801.