guard against LAPACK least squares solver test failure, round 2

[Sacha0]

#25287 modified a LAPACK least squares solver test to guard against failure due to poor conditioning of the test matrices. Unfortunately that modification proved insufficient as I ran into the motivating failure again locally

julia> srand(0xab6820ee372333e0a07585ed1a906b41);

julia> @testset "gelsy, gelsd" begin
           let elty = ComplexF64
               A = rand(elty, 10, 10)
               B = rand(elty, 10, 10)
               C, j = LAPACK.gelsd!(copy(A),copy(B))
               D, k = LAPACK.gelsy!(copy(A),copy(B))
               @test C ≈ D rtol=eps(cond(A))
           end
       end
gelsy, gelsd: Test Failed at REPL[34]:7
  Expression: ≈(C, D, rtol=eps(cond(A)))
   Evaluated: Complex{Float64}[-0.0262218+0.441148im 0.0346061-0.375153im … -0.0593423+0.645078im -0.166242+0.71145im; 0.392516+0.673879im 0.727514+0.518957im … 0.864918-0.22036im 1.0831+0.158312im; … ; 0.283946-0.078362im -0.161999-0.0885647im … -0.173693+0.390815im -0.474193-0.293889im; 0.41677-0.422742im 1.48165+0.0538742im … -0.153264-0.742447im 0.636379-0.666813im] ≈ Complex{Float64}[-0.0262218+0.441148im 0.0346061-0.375153im … -0.0593423+0.645078im -0.166242+0.71145im; 0.392516+0.673879im 0.727514+0.518957im … 0.864918-0.22036im 1.0831+0.158312im; … ; 0.283946-0.078362im -0.161999-0.0885647im … -0.173693+0.390815im -0.474193-0.293889im; 0.41677-0.422742im 1.48165+0.0538742im … -0.153264-0.742447im 0.636379-0.666813im] (rtol=3.552713678800501e-15)
Stacktrace:
 [1] macro expansion at ./REPL[34]:7 [inlined]
 [2] macro expansion at /Users/sacha/pkg/julia/usr/share/julia/site/v0.7/Test/src/Test.jl:980 [inlined]
 [3] top-level scope at ./REPL[34]:2
Test Summary: | Fail  Total
gelsy, gelsd  |    1      1
ERROR: Some tests did not pass: 0 passed, 1 failed, 0 errored, 0 broken.

This pull request relaxes the test tolerance just enough for the test to pass 100,000 consecutive repetitions:

julia> srand(0xab6820ee372333e0a07585ed1a906b41);

julia> @testset "gelsy, gelsd" begin
           @testset for elty in (Float32, Float64, ComplexF32, ComplexF64)
               for k in 1:10^5
                   A = rand(elty, 10, 10)
                   B = rand(elty, 10, 10)
                   C, j = LAPACK.gelsd!(copy(A),copy(B))
                   D, k = LAPACK.gelsy!(copy(A),copy(B))
                   @test C ≈ D rtol=4*eps(cond(A))
               end
           end
       end
Test Summary: |   Pass   Total
gelsy, gelsd  | 400000  400000
Test.DefaultTestSet("gelsy, gelsd", Any[Test.DefaultTestSet("elty = Float32", Any[], 100000, false), Test.DefaultTestSet("elty = Float64", Any[], 100000, false), Test.DefaultTestSet("elty = Complex{Float32}", Any[], 100000, false), Test.DefaultTestSet("elty = Complex{Float64}", Any[], 100000, false)], 0, false)

(Factors of 2 and 3 rather than 4 were insufficient.) Best!

[Sacha0]

(AV timeout seems unrelated.)

[Sacha0]

CI being happy, Fredrik having approved, Andreas having suggested the approach on slack, and this being a trivial change, absent objections or requests for time I plan to merge this pull request tomorrow midday PT at the earliest. Best!

[Sacha0]

Thanks all! :)