xla

Fix output type of custom calls while lowering quant/dequant torch op to HLO

#6283

Merged