Breaking up a common import cycle

[Joachim Durchholz]

During my attempts at C elimination, I see that atan2 is involved in
many potential import cycles.
Of these, most if not all can be broken up by splitting up
sympy.functions.elementary.trigonometric , e.g. by splitting off
sympy.functions.elementary.trigonometric_inverses.

Such a split is not a guarantee that it will prevent all present or
future atan2 cycles, because there's always the possibility of mutual
recursion between modules.
The alternate approach would be to use a local import, or a bottom import.

I'd prefer to split the module in this case, because it's a reasonable
module boundary anyway.
What do others think?