Pyheadtail

Changes:

- license: 3-clause BSD license
- wakes: adding vertical parallel plates, renaming horizontal ones
- longitudinal: adding square-well potential (ABS beams)
- feedback: ideal transverse damper now also includes reactive part (see [PyHEADTAIL-playground](https://github.com/PyCOMPLETE/PyHEADTAIL-playground/))
- monitors: fixing CellMonitor also for GPU (now used in production)
- wakes: bug fix in resistive wall scaling
- generators: bug fix in kv2D
- context management: fixed and new functions
- generators: function to accept generated particles only within certain action (useful especially for linear generators)
- setup.py: speeding up cython (removing profiling flag)

Changes:
- hotfix: CUDA 9.1 compatibility
(thrust::gather has been moved to thrust/gather.h)
- include cellmonitor (cf. https://github.com/PyCOMPLETE/PyHEADTAIL/pull/75)

Changes:
- bug fix for field map calculation in `FrozenGaussianSpaceCharge`
- `FrozenGaussianSpaceCharge` can recompute field map from beam
- new automated adaptive mode for `FrozenGaussianSpaceCharge`
- PIC / self-consistent space charge: same new automated adaptive mode implemented in `SpaceChargePIC_Adaptive25D`

Bugfixes and more flexibility for damper

- feedback: include damping rate 0 treatment
- frozen space charge field map: bug fix of vertical plane

New feature for FrozenGaussianSpaceCharge25D: providing adaptive mode

The fields are automatically recomputed when the beam RMS size changes. This is controlled via sigma_rtol.

Changes:

- Longitudinal RFBucket matching has been upgraded, support of thermal,
q-Gaussian, parabolic and waterbag distributions. Can easily be extended
to new distribution types. Make sure to check out the notebook
https://github.com/PyCOMPLETE/PyHEADTAIL-playground/RFBucket_Matching.ipynb .
- New FieldMap classes for 1D (horizontal), 2D (transverse) or 3D field maps
including the GPU (uses PyPIC).
- New FrozenSpaceCharge25D allows to apply a frozen transverse field map to a
bunch in slice-by-slice manner, where the longitudinal plane is self-consistently
resolved via slicing by using the local line charge density for the field map.
- New PyPIC version significantly speeds up self-consistent PIC space charge
on GPU (cf. message of commit 28e3cc5 for timing details).
- Self-consistent PIC space charge uses double precision by default now.
- 2D PDF integrators revived and restructured.
- RFBucket allows proper PS Booster accelerating bucket description now (cf. commits d49c460 and 2355c13).
- Release script includes PyPI publication now.
- Minor documentation improvements.
- Some (relevant) bug fixes.