disable multi_linear_share_same_input for dynamic shape case (#107123)
`reshape_linear_reshape_pattern` will fail for dynamic shapes and break the fusion.
We will disable this optimization for dynamic shapes, since the shape might be changed during runtime so we cannot compare the size hint.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107123
Approved by: https://github.com/XiaobingSuper, https://github.com/jgong5, https://github.com/jansel