Fix pivoted cholesky docstrings

[nalimilan]

The inverse permutation has to be used in the second equation, and the rank has to be taken into account for rank-deficient cases.
This doesn't make a difference in the simple example shown in the docstring, but it does in more complex cases.

(These docs were introduced by #39964)

[dkarrasch]

LGTM. How about actually adding a rank-deficient example? Something like

julia> X = [1.0, 2.0, 3.0, 4.0]; A = X*X';

julia> F = cholesky(A, Val(true), check = false);

julia> Ur = F.U[1:F.rank, :]; Lr = F.L[:, 1:F.rank]

julia> A[F.p, F.p] ≈ Ur' * Ur ≈ Lr * Lr'
true

? To reduce the doctests, one could remove the full-rank case, since that one is covered in the non-pivoted case already?

[nalimilan]

Good idea. Done.