I noted that the `right_prec` and `next_prec` functions taking a
character only work because prefix operators come after infix operators
in the table. We should compare the *actual* operator kind, which would
probably need a single operator enum to make sure there are no
collisions.
The previous version of the algorithm did not take into account the fact
that the main loop should only run while we have postfix or infix
operators. This happened to work because of the small amount of
operators used in the grammar.
The previous version also had prefix operators hard-coded in the operand
parsing function.
