Improve faulty parsing of assigments

[privat]

Add a specific handling on the parser when a lone := is found. It produces instances of a new specific class RBAssignmentErrorNode.

Bad left values, like in a foo := 5 are likely to be consumed as unfinished statements (like a foo.), then the := that follows (and the right value) consumed as distinct statements of assignment errors (like := 5). Thus, producing an acceptable faulty AST.

A possible alternative faulty AST could be to include the unfinished left value as a child of the RBAssignmentErrorNode.
But this requires to hack the parser more (for not that much additional value)