QR factorization's memory allocation when updating matrix A

[TongtongJin]

The original comments are:

We need to be careful that the factorization is no longer needed prior to updating A.

For example

A = randn(10, 4)
F = qr!(A)
R = F.R
A .= randn(10, 4)
R != F.R  # This should evaluate to true

If we do not do the in-place operation (i.e., solver.A = A) then something else will be allocated

A = randn(10, 4)
F = qr!(A)
R = F.R
A = randn(10, 4)
R == F.R # This should evaluate to true