Pynbody

What's Changed
* Support for swift, velociraptor and HBT+ (https://github.com/pynbody/pynbody/pull/757, https://github.com/pynbody/pynbody/pull/779, https://github.com/pynbody/pynbody/pull/780)
* Add standardised mechanism for getting all halo properties by apontzen in https://github.com/pynbody/pynbody/pull/777
* Make cuboid filters wrap around the box and improve implementation (https://github.com/pynbody/pynbody/pull/780)


**Full Changelog**: https://github.com/pynbody/pynbody/compare/v2.0.0-beta.3...v2.0.0-beta.4

What's Changed
* Fix a bug with reading gadget4-subfind-hdf tree information, which led to silent failures to load subfind catalogues
* Fixes a bug where tipsy files silently didn’t load if warnings were set to raise and there was no paramfile
* Reorganised the test data archive


**Full Changelog**: https://github.com/pynbody/pynbody/compare/v2.0.0-beta.2...v2.0.0-beta.3

Bug fix

Fixes getting halo properties in physical units (https://github.com/pynbody/pynbody/pull/775)

What's Changed

Near complete re-implementation of the halo catalogue class hierarchy (https://github.com/pynbody/pynbody/pull/763)

This is a _breaking change_, and as such the major version number of pynbody has increased. Regular users probably _don't_ want to install a v2 release at this early stage, and as such it has been flagged as a beta.

* Halo classes are no longer imported directly into the `pynbody.halo` namespace.
* Users may now request a particular halo finder, or list of halo finders, by passing a `priority` kwarg to `SimSnap.halos()`
* `HaloCatalogue.precalculate()` has been renamed to `HaloCatalogue.load_all()`
* Halo classes supporting child/parent relationships now expose a `.subhalos` property, in which the child halos are enumerated
* Behind the scenes, the mapping from `iords` to file offsets has been improved and unified across all `HaloCatalogue` subclasses
* `GrpCatalogue` has been renamed to `HaloNumberCatalogue` to better reflect its meaning
* `AHFCatalogue` no longer renumbers the halos using a one-based scheme by default. If AHF has written out halo numbers, these are used by default; if it has not written out halo numbers, a zero-based indexing is used. Users can pass `halo_numbers='v1'` to `halos()` to obtain the old behaviour
* The ability to run halo finders has been removed (it anyway only ever worked for AHF, and even then was a bit flakey)


**Full Changelog**: https://github.com/pynbody/pynbody/compare/v1.6.0...v2.0.0-beta

What's Changed

Significant enhancements to the KDTree (https://github.com/pynbody/pynbody/pull/769):
* pynbody's sphere filters now make use of the KDTree to greatly enhance extracting spheres from large simulations.
* KDTrees will not automatically be built, so you should call `f.build_tree()` before `f[pynbody.filt.Sphere(...)]` to obtain the speed-up. Building the tree takes a while, so this is only worth doing if a large number of spheres will be extracted.
* KDTrees are now built in parallel, significantly speeding up operations that need them (such as SPH smoothing)
* KDTree data is now stored in numpy arrays, so it can be inspected from python and also can be stored in shared memory

**Full Changelog**: https://github.com/pynbody/pynbody/compare/v1.5.2...v1.6.0

This corrects an error in the minimum python version (should be 3.9, not 3.8).

Versions 1.5.0/1.5.1 are incompatible with 3.8, but will incorrectly attempt to install on 3.8. They will therefore be yanked from PyPI.