Improve spliting into components

[lefessan]

I would advise to split the code in 4 parts:

• utils (already exists)
• parser : a library with the AST, a textual parser and a textual printer
• interpreter: a library to execute the AST over data
• niagara: a command-line application to test the compiler/interpreter on source files

Important parts:

• neither parser nor interpreter should contain code that displays textual messages. For example, they could define a error type the list all possible errors, instead of raising failwith or similar textual operations. They should also never call exit, but raise an exception instead.

[Keryan-dev]

Splitting done :

• commons for general data structures that have uses across the board
• surface for source language AST
• internal for intermediary representation
• grammar for source code parsing
• compiler for the compilation pipeline
• interpreter
• and exec for the CLI executable

Removing explicit errors from the libraries should be a separate issue.