Fixed EB data for anisotropic grids: boundary centroid, normal vector (backward compatible)

[skang67]

Summary

1. This PR corrects two EB data quantities: the boundary centroid (bcent) in set_eb_data called within build_cells, and the normal vector (normal) in EBToPVD::EBToPolygon.
2. This correction only affects anisotropic grids (dx != dy != dz) and does not affect computations for isotropic grids (dx==dy==dz).

Additional background

1. The current bcent, normalized using a single parameter ((nx*dx[0])*(nx*dx[0]) + (ny*dx[1])*(ny*dx[1]) + (nz*dx[2])*(nz*dx[2])) for all x-, y-, and z-components, results in inappropriate non-dimensional values. The proper physical coordinates can be calculated in user-end code (e.g. ERF) by re-normalizing the bcent parameters.
2. However, when constructing multilevel EB data by coarsening the finest EB data in Level::coarsenFromFine, the bcent from the finest grid is used to construct coarse EB data. This process produces incorrect bcent on coarse grids, which cannot be recovered in user-end code. Therefore, I propose this PR.

@nataraj2 I am submitting a PR to fix normalization of bcent for anisotropic grids. So far, I have used the bcent as it is, by renormalizing it in ERF. However, I encountered an issue that cannot be fixed from ERF when coarsening the EB data from the fine grid to coarse grids. I discussed this issue with @WeiqunZhang.

Checklist

The proposed changes:

• fix a bug or incorrect behavior in AMReX
• add new capabilities to AMReX
• changes answers in the test suite to more than roundoff level
• are likely to significantly affect the results of downstream AMReX users
• include documentation in the code and/or rst files, if appropriate

[nataraj2]

Sounds good.