For #67. Seems like this was overlooked in
ROCm/rocPRIM#734.

This change resolves a test failure observed in a future rocThrust test.
This is a blocker for adding CCCL 2.7 parity in rocThrust and should be
included in ROCm 7.0.

It adds 2 changes:

- Skip including rocprim_version.hpp when `_CLANGD` is defined. This
resolves `clangd` complaining about things.
- Fix incorrect masking in DPP warp scan.

---

Originally, executing the scan operator was not masked. That means that
all lanes could potentially execute. In the case of the failing thrust
test, it involved some pointer chasing. My guess is that some
optimization logic would still execute the scan operation (read: pointer
chasing) even though the input of the operation is optimized out. Since
we don't know what the values of those registers are when we do the scan
operation, it may chase the pointer to something in the middle of
nowhere, causing a crash.

The behaviour of hipGetLastError is changing in HIP 7.0. Previously the
error that was reported was cleared on each HIP API call. This means
that hipGetLastError reported any error that occurred during the last
HIP API call.

Moving forward, the error that's reported will only be cleared on each
call to hipGetLastError. This means that hipGetLastError will report any
error that has occurred since the last call to hipGetError.

Some of our tests rely on observing a return value of
hipErrorOutOfMemory from hipMalloc when an allocation is too large for a
given GPU architecture's memory system. This sets the internal HIP
error, and it's not cleared before subsequent tests call
hipGetLastError, causing them to fail.

This change adds extra calls to hipGetLastError to clear the error (for
future tests) in cases where tests run out of memory.

…135)

In the batch_memcpy_impl struct, we define two types based on the value
of the InputBufferItType template argument.
1. input_type, set to the InputBufferItType's underlying value_type's
value_type.
2. Alias, a type that's set using a std::conditional statement that
examines the IsMemCpy boolean template arg, and may be set to either
unsigned char or the type from (1).

Later code creates a reference to (1). This causes problems when (1) is
void.

This change removes the definition of (1), since it does not appear to
be used anywhere else in our repos (it's part of the detail namespace,
so it's not public).

However, this is not enough to fix the problem, since the compiler
evaluates both sides of the std::conditional we use to define (2). To
work around this, this change also adds a helper struct type,
(AliasType) that uses template specialization on the IsMemCpy boolean to
define the type that Alias will be assigned. This allows us to remove
the std::conditional statement.

This change did not seem to make it in this PR
ROCm/rocPRIM#727.

`detail::raw_storage` is still used quite often due to its performance
relative to other APIs. Let's undeprecate this until the alternatives
are equally performing.