Bernoulli polynomial problem

[davidloeffler]

The "bernoulli_polynomial" command should return the nth Bernoulli polynomial evaluated at the first argument, but something odd happens when you feed it a polynomial that's not a single variable:

sage: R.<x> = QQ[]
sage: bernoulli_polynomial(x, 2)
x^2 - x + 1/6
sage: bernoulli_polynomial(x - 1, 2)
x^2 - x + 1/6
sage: bernoulli_polynomial(R(0), 2)
x^2 - x + 1/6

[fwclarke]

In the attached patch I have rewritten most of the `bernoulli_polynomial` code.  This solves this problem.

In the case where the result is a polynomial, direct calculation from the list of coefficients is retained (for speed), but corrected for the case where x is not the variable. This only works when the base ring is a QQ-algebra; previously an error was raised it if wasn't.

A small increase in speed results from the odd Bernoulli numbers being zero, apart from the first. This is more significant in the non-polynomial case.

A couple of new doctests have been added.

[davidloeffler]

With the new patch, it not only gets the right answer, but it does so faster than it previously got the wrong answer. Can't argue with that.