[Quant][fx] Remove dequant-quant around getitem (#82675)
Summary: https://github.com/pytorch/pytorch/issues/82480 saw
unnecessary dequant-quant pairs around the getitem op, which led
to significant slowdowns. This commit simply removes this pair in
the lowering step, since getitem already handles quantized inputs.
Test Plan:
python test/test_quantization.py TestQuantizeFxOps.test_getitem_no_dequant_quant
Reviewers: jerryzh168
Subscribers: jerryzh168, supriyar
Tasks: https://github.com/pytorch/pytorch/issues/82480
Differential Revision: [D38427508](https://our.internmc.facebook.com/intern/diff/D38427508)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82675
Approved by: https://github.com/jerryzh168
