Condition number is computed wrong in some cases

[goggle]

I have discovered this issue while working on another one:

julia> M = [1.0 -2.0; -2.0 -1.5]
2×2 Array{Float64,2}:
  1.0  -2.0
 -2.0  -1.5

julia> S = Symmetric(M)
2×2 Symmetric{Float64,Array{Float64,2}}:
  1.0  -2.0
 -2.0  -1.5

julia> cond(M, 1)
1.909090909090909

julia> cond(S, 1)
2.227272727272727

These two condition numbers should be the same, but they are clearly not.

[antoine-levitt]

The issue is that _cond1Inf in lu.jl calls gecon, which is only an estimate of the matrix norm, so S is actually the correct one. Let's just remove the specialization and just compute opnorm(A, p)*opnorm(inv(A), p) in all cases? Condition number is only a diagnostic tool, so its speed is not that important. Affects #664.

[goggle]

@antoine-levitt Yes exactly, I was preparing a PR for #664, when this issue appeared after writing some tests for it.

Yes, I think removing the specialization

_cond1Inf(A::StridedMatrix{<:BlasFloat}, p::Real)

would resolve this.

[antoine-levitt]

I went ahead and did a PR for this, if you have any additional fixes or tests feel free to add them