Nearly all special errors are now handled gracefully, i.e. the
parser will be able to continue after encountering them. In some
cases the associated error range has been improved using the new
end attribute stack.
To achieve this the error handling code has been moved out of the
node constructors and into special methods in the parser.