Vegas

========================
- Integrator.random_batch() no longer exposes internal work buffers to users (it
now makes copies).

- Inconsistent behavior between adapt=False with minimize_mem=True
or with minimize_mem=False is now fixed.

- Cleaned up Cython.

======================
Modified so Integrator works with integrands that return 32-bit floats
as well as 64-bit floats. This is to accommodate integrands that are run on GPUs.

See updated documentation for examples that use GPUs, Cython, numba, or Pythran to
speed up vegas integrations.

Adds method Integrator.sample() for generating sample sets of integration points.

=========================
Modified so that it works with numpy>=2.0.

========================

- Fixes bug in implementation of PDFIntegrator's scale parameter. There was no
problem when scale=1 but other values were wrong.

- Fixes bug when calculating the number of degrees of freedom when averaging
results over multiple iterations with multiple integrands.

- Tweaks to PDFIntegrator.

========================
- More robust treatment of round-off errors when calculating means and variances in vegas.

=========================
- Adds PDFIntegrator.sample() which creates (weighted) samples
from the integrator's probability distribution (eg, for making
histograms or contour plots of probability distributions).

- Fixes defect in handling of round-off errors (in sigf).