Fix new IT blocks ARMv8

ARMv8 specifies that an IT block should be followed by only one 16-bit instruction.
* SkFloatToFix is back to a C implementation that mirrors the assembly code.

* S32A_D565_Opaque_neon switched the usage of the temporary 'ip' register to let
the compiler choose what is best in the context of the IT block. And replaced
'keep_dst' by 'ip' where low register or high register does not matter.

BUG=skia:

CQ_INCLUDE_TRYBOTS=skia.primary:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD

Change-Id: If587110a0c74b637ae99460419d46cf969c694fc
Reviewed-on: https://skia-review.googlesource.com/9346
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Mike Klein <mtklein@chromium.org>
2 files changed