Adaptive quadrature option

[unalmis]

• As discussed in some old pull requests, it could be useful to have an option to use adaptive quadrature for the bounce integrals. Since the fixed quadrature is limited by Patch for differentiable code with dynamic shapes #1303, the performance hit of an adaptive may not be that bad, since one could just check before applying the quadrature if the integrals has the same endpoints, and early exit before interpolating.

• This should be a few line changes in the bounce.integrate method to using the adaptive quadrature from quadax.quadgk. In my testing for this months ago, I determined the most efficient option was the sin change of variables with Gauss-Kronrod.