Fbpic

This release introduces an important change on GPU: FBPIC now uses the `cupy` package instead of the `pyculib` package (since `pyculib` is no longer supported); see [356](https://github.com/fbpic/fbpic/pull/356) [#363](https://github.com/fbpic/fbpic/pull/363) [#367](https://github.com/fbpic/fbpic/pull/367). As a result, it is now possible to use FBPIC with Numba 0.43 (or higher).

It also introduces official support for MPI-decomposed simulation ;
see [348](https://github.com/fbpic/fbpic/pull/348)

In addition, this release adds various improvements to FBPIC:

**New features:**

- FBPIC can now initialize beams with arbitrary charge [364](https://github.com/fbpic/fbpic/pull/364)
- The external fields can be automatically transformed to the boosted frame
[342](https://github.com/fbpic/fbpic/pull/342)
- The `ParticleChargeDiagnostic` is now smoothed, in the same way as the regular charge density diagnostic [349](https://github.com/fbpic/fbpic/pull/349)

**Optimizations:**

- The number of threads per block were optimized for modern GPUs [365](https://github.com/fbpic/fbpic/pull/365)
- Certain arrays are now kept on GPU, and never copied to CPU [361](https://github.com/fbpic/fbpic/pull/361)

**Bug-fix:**

- There was a bug when restarting a simulation involving ionization on GPU [360](https://github.com/fbpic/fbpic/pull/360)

This release introduces a few minor changes:

- It is now necessary to use Numba 0.42 (or lower), due to limitations in `pyculib`.
- The `BoostedFrameDiagnostics` are now renamed as `BackTransformedDiagnostics`; see [339](https://github.com/fbpic/fbpic/pull/339)
- The user can now pass a physical time (instead of a number of iterations) to the `ParticleDiagnostic` and `FieldDiagnostic`; see [336](https://github.com/fbpic/fbpic/pull/336)

This release makes several improvements to FBPIC:

New features:
- It is now possible for the `ParticleDiagnostic` to output the gathered
E and B field (on the macroparticles), as well as the particles' Lorentz factor
to the openPMD files ; see the docstring of `ParticleDiagnostic`, and
[330](https://github.com/fbpic/fbpic/pull/330),
[316](https://github.com/fbpic/fbpic/pull/316).

Miscellaneous:
- It is now possible to use FBPIC with Numba 0.42 ;
see [319](https://github.com/fbpic/fbpic/pull/319)
- The installation documentation now includes information for Summit, as
well as updated information for Lawrencium ; see
[322](https://github.com/fbpic/fbpic/pull/322)

Bug-fixes:
- There was a bug in MPI communications on GPU (in the rare case where all
particles of a given species are located in the guard cells), that typically
resulted in a `CudaAPIError`. This is now fixed ; see
[333](https://github.com/fbpic/fbpic/pull/333).
- Calling `add_gaussian_bunch` with erroneous parameters (e.g. such that
the bunch is not located inside the simulation box) used to produce `NaN`
in the simulation. This is now fixed ; see
[331](https://github.com/fbpic/fbpic/pull/331).
- When using custom smoothers for the charge/current, although the Maxwell
solver did take into this custom smoother, the initial space-charge calculator
still used the default one. This is now fixed ; see
[326](https://github.com/fbpic/fbpic/pull/326).
- The function `add_gaussian_bunch` could potentially generate electrons
with negative `gamma` (when initializing a bunch with a large energy spread).
This is now fixed ; see [325](https://github.com/fbpic/fbpic/pull/325)

This is a bug-fix release.

It deals with the fact that FBPIC happens to be incompatible with Numba version 0.42.
FBPIC will now raise an error when Numba 0.42 is detected, and the user
will be prompted to install Numba 0.41.

In addition, the code can now perform particle sub-sampling in the openPMD diagnostic.

This release introduces various improvements to FBPIC:

New features:
- It is now possible to tune the amount of smoothing applied on the
charge and currents produced by the macroparticles; see [268](https://github.com/fbpic/fbpic/pull/268).
- The documentation of the boosted-frame technique has been expanded. A
new function allows to easily compute the number of PIC iterations that
need to be performed in the boosted frame ; see
[294](https://github.com/fbpic/fbpic/pull/294).
- The user can now set the name of the folder where the checkpoints are saved ; see
[305](https://github.com/fbpic/fbpic/pull/305).
- The continuous injection of plasma (with moving window) is now more robust.
This can in particular suppress noise that some users might have observed in the
up-stream, continuously injected plasma, for very long simulations.

Bug-fixes:
- There was a bug in the laser antenna when using multiple CPUs/GPUs. This bug
has been fixed ; see [309](https://github.com/fbpic/fbpic/pull/309).
- The `FlattenedGaussianLaser` had bugs, esp. when initializing it out of focus.
This is now fixed ; see [305](https://github.com/fbpic/fbpic/pull/305) and
[ 299](https://github.com/fbpic/fbpic/pull/299).

This is bug-fix release.
It corrects a bug that was happening exclusively for *cubic shape* deposition,
when using the CPU (i.e. the bug does not occur on GPU) ; see [297](https://github.com/fbpic/fbpic/pull/297)
In addition, this release adds a safe-guard for the sign of the charge, for Gaussian beams [295](https://github.com/fbpic/fbpic/pull/295), and for the sign of the Galilean velocity [#293](https://github.com/fbpic/fbpic/pull/293).