1. The stack should descend, not ascend;
2. The stack should be pushed byte for byte; meaning, when pushing P or
A, those should consume one byte, not two;
3. The MSB should be pushed first when doing JSR and BRK, which makes
some sense if you were reading the stack from $0100 - $01FF.
We should have done this all along. We hadn't accounted for if orig >=
0x80 and result < 0x80, but at that point, it's kind of silly to have a
long complicated condition that we can satisfy simply with an XOR.
(Which is what the machine would do anyway.)
The previous method was not incorrect, but set16 already handles the
little-endian logic for us, and there's no need to do something bespoke
here that we then need to test for separately.
Regarding soft switches, we had several we should have been listening
for on both reads and writes, but were only doing so on writes; this is
now fixed.
Regarding statuses, we were incorrectly calculating both carry and
overflow. This should now be fixed, although some quick examinations of
disassembly output suggest there is something else amiss. Debugging will
continue shortly.
This also adds tests for those functions. We have, furthermore, removed
some redundant (and non-externed!) functions from apple2.mem.c which was
the old bank switching code that got moved to apple2.bank.c.
Previously, we just used whatever the segment was that got passed in.
But aside from switching the stack and zero page, ALTZP also changes
where bank-switchable memory is read from and written to; it's either
always aux memory (ALTZP is on) or always main memory (ALTZP is off).
There is one failing test, which I expect to fail at this point; that's
because we need to write a read/write map function for the stack and
zero page so that they use aux memory when BANK_ALTZP is on, and main
memory when not.
To allow this to work, we had to allow the CPU struct to record what the
last opcode/operand/address were, although in truth we only needed the
last address.
