Investigate alternative partitioning in tridiagonal solver

[msimberg]

DLA-Future/src/eigensolver/tridiag_solver/kernels.cu

Lines 638 to 643 in 6ac429e

	// The number of non-deflated values
	k = n - thrust::count(par, c_ptr, c_ptr + n, ColType::Deflated);
	// Partition while preserving relative order such that deflated entries are at the end
	thrust::stable_partition_copy(par, in_ptr, in_ptr + n, out_ptr, out_ptr + k,
	PartitionIndicesPredicate{c_ptr});

could potentially be done using a copy and then in-place partition instead. Investigate if it's faster than the current method.

This is low priority unless we find evidence that this kernel is a bottleneck.