Cleanups for psi and bug fix for exp-integral-e.

* Allow (exp-integral-e 1 z) to work.
* psi
  * Handle psi(1) specially.
  * Do a better job with cot(%pi*z) when z is an odd multiple of 1/2
    where cot is 0.
  * Fib bug in computing the number of terms when we try to float a
    complex.  Just float the realpart.