Use 3D Kokkos MDRangePolicy and remove manual SIMD loops in ComputeE/ComputeB functors

[Amazingkivas]

Motivation

• Replace manual i-range splitting and SIMD/unrolled loops with a full 3D Kokkos parallel policy to simplify the functors and improve portability across backends.
• Correct and consolidate operator signatures and bounds handling so each functor receives and iterates (k, j, i) and computes flat indices consistently.
• Fix parameter const-correctness for coefficient arguments to avoid accidental modification and clarify intent.

Description

• Removed start_i/end_i members and all SIMD/unrolled loop code paths from ComputeE_FieldFunctor and ComputeB_FieldFunctor, and replaced them with an element-wise implementation using a 3D Kokkos::MDRangePolicy.
• Updated apply functions to create a 3D MDRangePolicy (with explicit iteration order) using {bounds_k[0], bounds_j[0], bounds_i[0]} -> {bounds_k[1], bounds_j[1], bounds_i[1]} and adjusted parameter lists accordingly, including making coef_dz a const FP&.
• Changed operator() signatures to operator()(const int& k, const int& j, const int& i) const, computed the flattened index, and replaced the SIMD neighbor-index logic with scalar neighbor index calculations (e.g. i_pred_idx, j_pred_idx, k_pred_idx) and periodic boundary handling.
• Applied the same scalarization and indexing simplifications to ComputeB_FieldFunctor so both field-update functors use the same 3D policy and indexing approach.

Testing

• Built the project and ran the automated unit and integration tests that exercise the Kokkos functors and field-update kernels, and the test suite completed successfully.
• Ran the project's field update regression tests under the modified functors and they passed with no failures.

---

Codex Task