Loosen test of Cholesky factorization of diagonal matrices to approxi…

…mate (JuliaLang#34859)

equality since BLAS and Julia might use square roots that round differently.
This has occasionally caused CI failures.

stdlib/LinearAlgebra/test/cholesky.jl

@@ -338,7 +338,7 @@ end
  CD = cholesky(D)
  CM = cholesky(Matrix(D))
  @test CD isa Cholesky{Float64}
- @test CD.U == Diagonal(.√d) == CM.U
+ @test CD.U ≈ Diagonal(.√d) ≈ CM.U
  @test D ≈ CD.L * CD.U
  @test CD.info == 0
 
@@ -352,7 +352,7 @@ end
  CD = cholesky(D)
  CM = cholesky(Matrix(D))
  @test CD isa Cholesky{Complex{Float64}}
- @test CD.U == Diagonal(.√d) == CM.U
+ @test CD.U ≈ Diagonal(.√d) ≈ CM.U
  @test D ≈ CD.L * CD.U
  @test CD.info == 0