PySDM v3.0.0-pre.4

What's Changed

• CI: workaround failing Homebrew Paraview downloads by @slayoo in #1692
• global backend instance cache by @slayoo in #1691

starting from this release, backend instantiation works as follows (with the API essentially unchanged):

• Numba(formulae=None, ...) and ThrustRTC(formulae=None, ...) return a new backend instance each time called
• CPU(formulae=None, ...) and GPU(formulae=None, ...) use a global backend instance cache (with formulae and all other **kwargs featured in the cache key)

Thus, any code that used the CPU/GPU aliases before, now uses the new cache. This provides a considerable speed up in cases where multiple simulations are performed, and where previously the backend was not shared among particulators.

Full Changelog: v3.0.0-pre.3...v3.0.0-pre.4