mul! returns invalid result for PDiagMat

[oschulz]

LinearAlgebra.mul! returns an invalid result for PDiagMat. Given

using LinearAlgebra, PDMats

A = PDiagMat([0.25, 0.75])
x = [0.0, 1.0]
y = similar(x)

We get

A * x == [0.0, 0.75]

as expected, but mul! gives

mul!(y, A, x) == [0.0, 0.0]

However,

mul!(y, Matrix(A), x) == [0.0, 0.75]

returns the correct result.

For some reason, LinearAlgebra.generic_matvecmul! doesn't seem to work correctly for PDiagMat.

Tested with Julia version v1.5.3 on Linux x86_64.

CC @vindex10, who noticed incorrect results caused by this in a actual use case.

[oschulz]

Thanks!