Fastmbar

**Bug fixes:** A bug was discovered and fixed. The bug affects the calculation of DeltaF_std when Bootstrap is used.

This is a major update of FastMBAR. FastMBAR is entirely rewritten to be faster, cleaner, and more concise. In addition, the following new features are added.

**New features**
- Newton's method is added to solve the MBAR equation.
- Asymptotic variance will be computed to estimate the uncertainty when Bootstrap is not used.

**Bug fix:**
np.int is deprecated in numpy 1.20.0, so np.int is replaced with the builtin int to silence the deprecation warning.

1. With this version, the size of MBAR/UWHAM equations that can be solved by FastMBAR is not restricted by the memory size of a GPU. This means FastMBAR can a GPU to solve MBAR/UWHAM equations where the size of the energy matrix is larger than the memory size of the GPU.

2. Documentations for important attributes and methods of the FastMBAR class were added.