use LinearAlgebra's new generic in-place 5-arg mul! wherever possible (requires Julia 1.3)

[chriscoey]

see JuliaLang/julia#29634

this will fix #206, #245 and is relevant to #251, #248

[davidschlegel]

I think this will also fix #295 .

[ranocha]

I think #295 will also require updates of lines like

IterativeSolvers.jl/src/bicgstabl.jl

Lines 37 to 40 in ae01dfe

	# Large vectors.
	r_shadow = rand(T, n)
	rs = Matrix{T}(undef, n, l + 1)
	us = zeros(T, n, l + 1)

[davidschlegel]

Yes, and probably at some other lines. I am not familiar with the code, so I think I cannot really help here, but I would love to have this fixed. The cg function already has the functionality to work with CUDA types, and so I think it should be possible to do it here are well. #295