Add approximate location to recovered token

Copied from commit message:

Problem: now synthesized by error recovery tokens have ghost locations that can be confusing

Solution: remember the last token that the parser consumed and use its end as the start of the synthesized token. So it gives a good approximation of the potential location for missing tokens.

Other improvements:

• align recovery annotations in the ParToken.mly
• fix pretty printing of module alias in the pascaligo