Add helper method for efficiently combining logical operations

[publixsubfan]

In #577, it was discovered that using bitwise AND to combine the results of single-dimension checks was considerably faster than using logical AND on the GPU:

bool status = true;
for (int idim  = 0; idim < NDIMS; idim++)
{
    // generates predicated branches: slower on GPU, faster on CPU
    status = status && detail::intersect(...);
    // faster on GPU
    status = status & detail::intersect(...);
}

Conversely, the logical AND seems to be faster than the bitwise AND on the CPU.

We should create a helper macro/method that can pick between bitwise and logical AND depending on whether the code is being compiled for the CPU or the GPU.

[kennyweiss]

There are open questions about how to best deal with this.

• Macros are likely the most performant way to deal with this, but not the most readable.
• Inline functions might not generate optimal assembly code

[kennyweiss]

Perhaps add this to a "benchmarks" suite to track performance across platforms?

[kennyweiss]

See changes from #1171
Specifically https://github.com/LLNL/axom/pull/1171/files#diff-14a178f0280c77d0aaf9bb588a55616dcf7fd84651c3bb7584bd524dd70f3ae5