Fbpic

This is a bug-fix release. It ensures that threading is only used with the
proper numba version (numba 0.34). It also fixes some issues with the MPI
implementation (esp. in particle bunch initialization and charge conservation).

This is a bug-fix release, that solves an issue when using
openPMD-viewer >= 0.6.0 to read the checkpoint files (for a restart simulation).

This is a bug-fix release, that solves an issue with the particle
boosted-frame diagnostics, in the case where both the GPU and a moving window
were used.

This version brings two majors changes to FBPIC:

- The code now supports **multi-threading** on CPU (when using numba>=0.34),
and is therefore much faster then it used to be on multi-core CPU architectures.

- The code does not rely on the proprietary library `accelerate` anymore, and
uses the open-source library `pyculib` instead. As a consequence, **all** the
dependencies of FBPIC are now open-source.

In addition to these changes, several minor improvements have been made to the
GPU code, including faster sorting routines (`prefix_sum`) and shorter
compilation time (function signatures have been removed).

This is a bug-fix release, to solve an issue with the particle boosted-frame
diagnostics, in the case where the GPU is used. The particle
boosted-frame diagnostics now correctly run on the GPU.

This version incorporates ionization (ADK model). The implementation is
Lorentz-invariant and thus works in the boosted-frame. The implementation
is also fully compatible with GPU, MPI, openPMD diagnostics (including
boosted openPMD diagnostics), and tracking (e.g. of the produced electrons)

In addition, several improvements were made to the code in general:
- External bunches can now be loaded to the simulation from openPMD files,
or from numpy arrays.
- Particle tracking is now compatible with the boosted openPMD diagnostics.
- The laser can now be injected in the simulation with a temporal chirp.