Added SatWidenMulPairwiseAccumulate and SatWidenMulAccumFixedPoint ops

[johnplatts]

Resolves issue #2050

The SatWidenMulPairwiseAccumulate(DI32, VI16 a, VI16 b, VI32 sum) op was added as AVX3_DL/PPC8/PPC9/PPC10 can carry out the SatWidenMulPairwiseAccumulate(di32, a, b, sum) op using a single instruction.

Adding the SatWidenMulPairwiseAccumulate op also allows the SatWidenMulAccumFixedPoint(di32, a, b, sum) op to be more efficiently implemented for x86 and PPC targets.

On NEON, SatWidenMulAccumFixedPoint(di32, a, b, sum) is a wrapper around the NEON vqdmlal_s16 op.

Similarly, on SVE2, SatWidenMulAccumFixedPoint(di32, a, b, sum) is a wrapper around the SVE2 svqdmlalb_s32 op.

[jan-wassenberg]

Very nice, thanks for adding this already!
FYI a compiler issue is causing our internal CI to fail. Will investigate.

[Ryo-not-rio]

Thank you for getting on this so quickly! Would it be possible to add the same for 8 bits and 16 bits as well?

[jan-wassenberg]

Thanks for adding this :) Are you also interested in extending it to other lane types?