[caffe2] remove quant options of SparseAdagrad from OSS (#35608)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35608
Various quantization options in SparseAdagradOp is only for experimental purposes and was unnecessarily complicating the code.
Moving these options back to internal code, merge into SparseSimdAdagradStochasticQuantOp, and change the name to SparseSimdAdagradFakeQuantOp
Test Plan: CI
Differential Revision: D20720426
fbshipit-source-id: 34c8fdea49f239c795f63e978ab13c8f535609d2