Fix DynamicView::resize_serial

[masterleinad]

I was running into a case where the call operator for DynamicView would deadlock in

kokkos/containers/src/Kokkos_DynamicView.hpp

Lines 435 to 438 in 902a50d

	// Allocation of this chunk is in progress
	// so wait for allocation to complete.
	while (nullptr == *ch)
	;

. As discussed on slack, the code there was only useful before the removal of resize_parallel in 768f657.
In particular, it's just an error if the chunk pointer we need to access is a nullptr. I took the freedom to remove all that logic in 6b7a1a0 to only check if for out-of-bounds errors and not checking the pointer to be a nullptr at all.

The origin of the issue was seeing was that the deep copy in resize_serial hadn't finished before we were trying to access the View. Conceivably, this can happen if the access happens on a different execution space instance than the one used for deep copy. In this case, it was the same execution space instance but the deep copy wasn't properly enqueued (also see #5065). Since we are supposed to fence after calling DeepCopy internally anyway, I didn't bother with adding the workaround there as well.
In the end, it was enough to just fence the deep copy operation in resize_serial, and to do that I had to provide ChunkManager::deep_copy_to with an execution space instance argument. This also allowed (and enforced) using an execution space instance in a couple of other places using deep_copy_to.

[masterleinad]

OPENMPTARGET-Clang failing with

4: [ RUN      ] openmptarget.reducers_struct
4: /var/jenkins/workspace/Kokkos/core/unit_test/TestReducers.hpp:367: Failure
4: Expected equality of these values:
4:   sum_view_scalar
4:     Which is: 32-byte object <00-B6 51-47 00-B6 51-47 00-60 D7-46 00-C0 58-46 00-B6 51-47 00-B6 51-47 00-B6 51-47 00-B6 51-47>
4:   reference_sum
4:     Which is: 32-byte object <00-B6 51-47 00-B6 51-47 00-B6 51-47 00-B6 51-47 00-B6 51-47 00-B6 51-47 00-B6 51-47 00-B6 51-47>
4: [  FAILED  ] openmptarget.reducers_struct (4 ms)

in openmptarget.reducers_struct looks unrelated/spurious.

[crtrott]

Ok this integral constant thing is the approved dummy object ...