Fix float vector comparisons with signed zero and NaN, add test coverage

[kcbanner]

Closes #21932

• Value: fix comparison of NaN in compareHeteroAdvanaced
• Sema: fix equality comparison of signed zeroes and NaN in compareScalar
• tests: add test coverage for vector float comparisons

[kcbanner]

I observed these failures on the CI, so I've disabled these new tests on these targets:

     +- run test behavior-arm-linux.4.19...6.11.5-eabihf-baseline-Debug 1871/1977 passed, 1 failed, 105 skipped
     +- run test behavior-arm-linux.4.19...6.11.5-musleabihf-baseline-Debug-libc 1871/1977 passed, 1 failed, 105 skipped
     +- run test behavior-thumb-linux.4.19...6.11.5-eabihf-baseline-Debug 1868/1977 passed, 1 failed, 108 skipped
     +- run test behavior-thumb-linux.4.19...6.11.5-musleabihf-baseline-Debug-libc 1868/1977 passed, 1 failed, 108 skipped
     +- run test behavior-powerpc64-linux.4.19...6.11.5-none-ppc64-Debug 1863/1977 passed, 1 failed, 113 skipped
     +- run test behavior-powerpc64-linux.4.19...6.11.5-musl-ppc64-Debug-libc 1863/1977 passed, 1 failed, 113 skipped
     +- run test behavior-powerpc64le-linux.4.19...6.11.5-gnu.2.28-ppc64le-Debug-libc
         +- zig test Debug powerpc64le-linux-gnu failure
error: Assertion failed: N->getValueType(0) == MVT::v1i1 && "Expected v1i1 type" (/home/andy/dev/zig-bootstrap/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp: ScalarizeVecOp_VSETCC: 913)

[jacobly0]

#22744 and maybe #22636

[andrewrk]

Rebased and force-pushed. The only missing consideration I see here has to do with comptime function call memoization.

comptime call memoization should only occur if the float values are bit-for-bit equal, since floats have well-defined memory layout.

I'm unsure whether this changeset respects that rule or not, and unsure whether there is already behavior test coverage for this or not.

[alexrp]

Breakage from changes in #22589: /home/ci/actions-runner9/_work/zig/zig/test/behavior/floatop.zig:147:52: error: no field or member function named 'floatAbi' in 'Target'

[kcbanner]

I'll update this PR this weekend - I'm not familiar with how to test for memoization behaviour but I assume there are existing tests I can reference for that.

[kcbanner]

Fixed up the breakage from #22589 and added comptime memoization tests.

[andrewrk]

Thanks!