Zabha and Zacas stay enabled after software clears misa.A

[akifejaz]

misa_csr_t::unlogged_write() re-gates EXT_ZAAMO and EXT_ZALRSC whenever software writes misa, so the base atomics correctly become illegal once misa.A is cleared. EXT_ZABHA and EXT_ZACAS should be added to that list I believe.

A byte or halfword AMO, or a compare-and-swap, therefore keeps executing after misa.A reads zero.

I cannot share the test (its long ) but these are the relevant prints.

misa before 0x8000000000141101  A=1
misa after  0x8000000000141100  A=0
control OK: misa.A really is clear
amoadd.b returned 0x0000000000000000, cell is now 0x0000000000000001
BUG: amoadd.b executed with misa.A clear
exit=1

while the expected (I believe) should be "IllegalInstruction (mcause = 2)".

[aswaterman]

Resolved by #2388

@akifejaz for future reference, if you write "Fixes #whatever" or "Resolves #whatever", it then merging the PR auto-closes the issue, but the extra "the" in your PR message prevented that trick from working.