parameterise gradient test for PD bijector properly instead of using

ForwardDiff as per suggestion of @devmotion

test/ad/pd.jl

@@ -1,16 +1,20 @@
+_topd(x) = x * x' + I
+
 @testset "AD for PD bijector" begin
  d = 4
  dist = Wishart(4, Matrix{Float64}(Distributions.I, d, d))
- x = rand(dist)
  b = bijector(dist)
  binv = inverse(b)
+
+ z = randn(d, d)
+ x = _topd(z)
  y = b(x)
 
- test_ad(vec(x); use_forwarddiff_as_truth=true) do x
- sum(transform(b, reshape(x, d, d)))
+ test_ad(vec(z)) do x
+ sum(transform(b, _topd(reshape(x, d, d))))
  end
 
- test_ad(y; use_forwarddiff_as_truth=true) do y
+ test_ad(y) do y
  sum(transform(binv, y))
  end
 end