Allow QR decomposition for zero-row matrices.

[manuelbb-upb]

This came up in DynareJulia/FastLapackInterface.jl#39

Currently, Julia allows for QR decomposition of zero-column matrices by skipping the call to LAPACK with a simple conditional [in this line].(https://github.com/JuliaLang/julia/blob/923fe2d74d469d73f0285a9cd41c88f90edd04d1/stdlib/LinearAlgebra/src/lapack.jl#L456)

Zero-row matrices are not supported and give an argument error:

 LA.qr(zeros(0, 10))

 ** On entry to DGEQRT parameter number  3 had an illegal value
ERROR: ArgumentError: invalid argument JuliaLang/julia#3 to LAPACK call

Is there a reason for this?
The data type should support these matrices:

import LinearAlgebra as LA
A = Matrix{Float64}(undef, 0, 10)
T = Matrix{Float64}(undef, 0, 0)
qr = LA.QRCompactWY(A, T)

and qr is as I would expect it.

[dkarrasch]

Would it fix the issue if we changed the condition to

n > 0 && m > 0

so that we go down the LAPACK path only in the LAPACK-valid cases ? If yes, would you be able to submit a PR, together with unit tests?

[manuelbb-upb]

That would indeed be sufficient. I will prepare a PR until tomorrow evening.

Edit: Will habe to delay for one or two days, other things got in the way. This line has stood for 11 years, so it is not too urgent, I guess 😅