#5694 elliptic_pi loses accuracy for m > 1

[Raymond Toy]

Claude discovered that elliptic_pi loses accuracy for m > 1.

For example, elliptic_pi(.5,2,2) returns
1.5327383516420856 - 2.9715954364976005*%i

The actual answer is
1.5327383490292811 − 2.971595438124559*%i

This was computed by Claude using mpmath with 300 digits. We only have about 8 digits correct.

[Raymond Toy]

Fixed by [6617c2] and following commits for the tests.