Try to autodetect the number of devices

[masterleinad]

Fixes #2975. After talking to @crtrott this is enough at a first step.
We might want to check that all the devices also satisfy the compute capability given or something similar.

So far, I checked that we arrive at the line changed in this pull request with -1 by default.

While there, I also changed the type of the numerous *_found variables to bool. I can rip that out into a new pull request if preferred.

[masterleinad]

Retest this please.

[masterleinad]

Retest this please.

[dalg24]

Why did you implement it here rather than Kokkos_HIP_Space.cpp?

[masterleinad]

HIPInternalDevices is declared in this file only in this file in an anonymous namespace. If preferred, we can also declare that class in Kokkos_HIP_Instance.hpp and move the implementation to Kokkos_HIP_Space.cpp.

[dalg24]

Fair enough. We will need to clean that out at some point. I am not asking you to change.

kokkos/core/src/HIP/Kokkos_HIP_Space.cpp

Lines 590 to 593 in 266724d

	int HIP::concurrency() {
	auto const& prop = hip_device_prop();
	return prop.maxThreadsPerMultiProcessor * prop.multiProcessorCount;
	}

kokkos/core/src/HIP/Kokkos_HIP_Space.cpp

Lines 631 to 633 in 266724d

	hipDeviceProp_t const& HIP::hip_device_prop() {
	return Impl::HIPInternal::singleton().m_deviceProp;
	}