ca65: HuC6280 zero-page is at $2000 and not at $0000

[jbrandwood]

CA65 currently assumes that zero-page/direct-page references resolve to absolute address $0000-$00FF.

This isn't correct on the HuC6280 where zero-page/direct-page references resolve to absolute addresses $2000-$20FF.

This leads CA65 to generate inefficient/wrong code for instructions referencing addresses $0000-$00FF and $2000-$20FF on the HuC6280.

I'm adding a pull request for a fix to this that differentiates between instructions with one-byte immediate operands that are always $0000-$00FF (handled by the existing old Emit1() function), from instructions that have one-byte direct-page/zero-page references where the top-byte of the address is different on different processors (handled by the new EmitZP() function).

EmitZP() is just a copy of the existing Emit1() function but with the check for the top-byte changing depending upon the currently-selected processor.

[jbrandwood]

Added pull request #318

[jbrandwood]

Withdrawn the pull request.

Travis CI has shown an unintended side-effect of CA65's expression handling that this patch trips up on.

The issue remains, but the "fix" needs more thought.

[jbrandwood]

Added new Pull Request #319, with notes on the fix.

The only "safe" solution that I've been able to come up with causes the HuC6280 to assemble zero-page instructions for both $0000-$00FF and $2000-$20FF.

This isn't perfect, but it is acceptable because the PC Engine doesn't ever need to access $0000-$00FF if the VDC is accessed through one of its mirror locations in $0100-$03FF.