Try to improve formula selection in acb_hypgeom_beta_lower

[fredrik-johansson]

This is not always better than the old code, but it does at least fix #1979.

The test code given in that issue now outputs 1.000000000 +/- 1.1293e-56.