fmin_powell can fail for out-of-bounds initial guess

[mmckerns]

>>> import mystic as my
>>> import mystic.models as mm
>>> rosen = mm.rosen
>>> bounds = [(0.0001, 0.0002)]*3
>>> _eqn = 'x0 + x1 + x2 <= .0005'
>>> eqn = my.symbolic.simplify(_eqn)
>>> cons = my.symbolic.generate_constraint(my.symbolic.generate_solvers(eqn), join=my.constraints.and_)
>>> guess = [0.001,0.001,0.001]
>>> tight = None
>>> solution = my.solvers.fmin_powell(rosen, guess, maxiter=50, bounds=bounds, constraints=cons, tight=tight)
/Users/mmckerns/lib/python3.7/site-packages/mystic-0.3.9.dev0-py3.7.egg/mystic/scipy_optimize.py:655: RuntimeWarning: invalid value encountered in subtract
  if (fx2 - fval) > delta:
/Users/mmckerns/lib/python3.7/site-packages/mystic-0.3.9.dev0-py3.7.egg/mystic/scipy_optimize.py:700: RuntimeWarning: invalid value encountered in subtract
  if (fx2 - fval) > delta:
Optimization terminated successfully.
         Current function value: inf
         Iterations: 2
         Function evaluations: 0
>>> print(solution)
[-10.35161585   5.17605792   5.17605792]

For the same problem, other solvers perform much better -- and at least don't fail to produce a single good result.