[PT2][Quant] Move to module partitioner for linear pattern quantization (#101122)
Subgraph matcher is somewhat unreliable as the pattern can vary depending on
the dimensionality of input tensor used to trace _and_ what appears before
linear
Differential Revision: [D45713915](https://our.internmc.facebook.com/intern/diff/D45713915/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101122
Approved by: https://github.com/jerryzh168
