[Quant][Inductor] Fix q/dq per channel lowering with 64-bit qparams (#120984)
Fixes #120869
Fix lowering of `quantize_per_channel` and `dequantize_per_channel` with float64 scale and int64 zero point.
Generated codes are incorrect without explicit type conversion. Add type conversion to the lowering pass, i.e., float64 (double) -> float32 and int64 -> int32.
**Test plan**
python test/inductor/test_cpu_repro.py -k test_per_channel_fake_quant_module_uint8
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120984
Approved by: https://github.com/leslie-fang-intel, https://github.com/jgong5, https://github.com/jerryzh168