DynRankView: make layout() return the same as a corresponding static View

[PhilMiller]

No description provided.

[PhilMiller]

Replaces #4775

[PhilMiller]

This got opened because I needed to move away from branches in the NGA repository to a personal fork, so maintainers can push changes on PRs if needed. The comments on the predecessor PR indicate why this shouldn't be merged as is.

[PhilMiller]

#5188 potentially unblocks this

[PhilMiller]

This test code that I needed to pass now passes:

    Kokkos::View<double****> u("foo", 1, 2, 3, 4);

    auto ulayout = u.layout();
    ASSERT_EQ(ulayout.dimension[0], 1);
    ASSERT_EQ(ulayout.dimension[1], 2);
    ASSERT_EQ(ulayout.dimension[2], 3);
    ASSERT_EQ(ulayout.dimension[3], 4);
    ASSERT_EQ(ulayout.dimension[4], KOKKOS_INVALID_INDEX);
    ASSERT_EQ(ulayout.dimension[5], KOKKOS_INVALID_INDEX);
    ASSERT_EQ(ulayout.dimension[6], KOKKOS_INVALID_INDEX);
    ASSERT_EQ(ulayout.dimension[7], KOKKOS_INVALID_INDEX);

[PhilMiller]

retest this please

[masterleinad]

Why is it advantageous for DynRankView to return the layout of the corresponding View instead of what it does currently?

[PhilMiller]

It allows easy construction of a View or DRV with matching rank and extent, and thus deep_copy between them.

[PhilMiller]

Also, generally, it's weird that the rank of the DRV doesn't match the rank described by its layout

[masterleinad]

Yes, I think this makes more sense than always returning a 7-dim layout.