mp.asin(z).imag and fp.asin(z).imag use inconsistent sign choices when z.real > 1 and z.imag==0

As in the title. Reproducer:

```python
>>> mpmath.mp.asin(2)
mpc(real='1.5707963267948966', imag='-1.3169578969248168')
>>> mpmath.fp.asin(2)                     # expected result
(1.5707963267948966+1.3169578969248166j)
```
