Allow two or more predecessor values to be passed to cuda_scheduler bulk customization

[msimberg]

Currently only a single or zero values are allowed to be passed from the predecessor sender (

pika/libs/pika/async_cuda/include/pika/async_cuda/cuda_scheduler_bulk.hpp

Lines 126 to 129 in d48a0d5

	// TODO: This should pass through the additional arguments.
	// Currently only works with zero or one arguments with a plain call
	// operator like here.
	void operator()(whip::stream_t stream) const { launch_bulk_function(f, shape, stream); }

). I think this could easily be generalized by collecting arguments into a tuple, and then using the unpack sender adaptor to split values again after the bulk operation.

Note, this is a good first issue if one is already familiar with CUDA. The changes required for this should be small, but may require a bit of background knowledge of CUDA and std::execution.