To run a parseExpr on first pass would always return a false success
with dummy value because symbols are configured to always succeed on
first pass. This would make expressions like ".fill 0x38-$" so bad
things to labels because "0x38-$" wouldn't return the same thing on
first and second pass.
Revert to parsing literals and symbols after having scanned for
expressions and add a special case specifically for char literals (which
is why we scanned for literals and symbols first in the first place).
Instead of buffering input in memory one line at a time, we go in "just
in time" mode and always read contents directly from I/O, without
buffering.
It forces us to implement a `ioPutback` scheme, but on the other hand it
greatly simplifies cases where multiple tokens are on the same line
(when a label is directly followed by an instruction).
The end result feels much more solid and less hackish.
