[llvm] [polly] [DA] Add batch delinearization support for improved precision (PR #170519)

[Ryotaro Kasuga via llvm-commits]

https://github.com/kasuga-fj commented:

- DA and Delinearization are already tightly coupled. I think it's better to separate them properly first. Maybe adding a feature like `BatchDelinearization` to Delinearization (like `BatchAliasAnalysis`) would be a good idea.
- Supporting both fixed-size and parametric batch delinearization at the same time makes the change complex. I think it's better to first support only fixed-size batch delinearization.
    - Notably, due to the recent changes in the validation logic of delinearization, currently parametric delinearization will fail almost all cases. This is because we cannot prove that the product of estimated dimensions (e.g., `N * M` for `A[N][M]`) doesn't overflow. See ##169048 and #169902 for details.

https://github.com/llvm/llvm-project/pull/170519