Make tests compatible with current numpy/scipy/pandas

Make tests compatible with current numpy/scipy/pandas

These issues are all within test code, i.e. should not affect users.

https://github.com/statsmodels/statsmodels/commit/bb8fba9d4d9883be2991b43b29bfa3a9c3fb325a
https://github.com/statsmodels/statsmodels/commit/310f91da9f07d9483965d624d76dbb7abebbd67f
https://github.com/statsmodels/statsmodels/pull/4911/commits/18dcceefbfed61b0d11ae5884a9f4cea82c2edb5
https://github.com/statsmodels/statsmodels/commit/f1d5eddd44d61099a9f8f90c40c35716a8346cf7

and one new fix only needed on 32 bit Python 2 (integer data loads as
int64, which gets converted to long producing factor names like
C(agecat)[T.4L], which does not match C(agecat)[T.4])

Author: Kevin "bashtage" Sheppard, Peter "thequackdaddy" Quackenbush, Matthew Brett, Rebecca N. Palmer <rebecca_palmer@zoho.com>
Origin: upstream (mostly)
Forwarded: no

Gbp-Pq: Name test_fixes.patch