fix vecnorm for sparse matrices

[Sacha0]

Reduced from the failures in #17507. Failure and fix demo:

julia> @test (foo = speye(4); resize!(foo.nzval, 5)[5] = NaN; vecnorm(foo) == 2.0)
Test Failed
  Expression: begin
    foo = speye(4)
    (resize!(foo.nzval, 5))[5] = NaN
    vecnorm(foo) == 2.0
end
ERROR: There was an error during testing

julia> Base.LinAlg.vecnorm(A::SparseMatrixCSC, p::Real=2) = vecnorm(view(A.nzval, 1:nnz(A)), p)

julia> @test (foo = speye(4); resize!(foo.nzval, 5)[5] = NaN; vecnorm(foo) == 2.0)
Test Passed

Best!

[andreasnoack]

Great. I fear that there might be a few more of these around. Maybe we should NaN initialize memory. Then it would be easier to find these cases.

[andreasnoack]

Please use multiple lines

[Sacha0]

Cheers, I will extend this test to multiple lines and then merge. Thanks!

[StefanKarpinski]

Maybe we should NaN initialize memory.

This is not completely crazy for an opt-in testing mode.

[Sacha0]

This is not completely crazy for an opt-in testing mode.

👍 Such a testing mode would have made diagnosing this issue much easier. Without a NaN serendipitously popping into the relevant uninitialized data on one run, chances are I would have been much slower to pin down the failure.