[VectorCombine] foldInsExtFNeg -support mismatching extract/insert indices

[RKSimon]

foldInsExtFNeg is currently limited to insert(dst, fneg(extract(src,idx)), idx) patterns with the same extract/insert index.

We should be able to handle cases where the (legal) indices are different with suitable shuffle masks + costs.

CC @ParkHanbum who's handled similar cases in the past

[ParkHanbum]

I would be happy to be able to improve what I did. Thank you!