Poppunk

The first bug fix will affect many results, and all users are encouraged to upgrade

New features:
- More thorough sample QC using pp-sketchlib features (101)
- Update to pp-sketchlib v1.5.1 (104)

Bug fixes:
- Misordered labels with older versions of pp-sketchlib (95)
- `TypeError` with visualisations (99)
- networkx still used in reference prune program (97)

- Update to sketchlib 1.4.0 (90) (https://github.com/johnlees/pp-sketchlib/releases/tag/v1.4.0)
- Use faster, threaded matrix functions (78, 80)
- Change from networkx to graph-tool (83)
- Use sharedmem (76)
- Add `--lineage-clustering` mode (72)
- Better `--refine-model` default boundary (94)

NB python >=3.8 is now *required* (81, 76)

This accommodates updates to pp-sketchlib introduced since v1.1.0, mostly passing through new command line parameters (e.g. `--use-gpu`). See https://github.com/johnlees/pp-sketchlib/releases/tag/v1.3.0 for more info.

This is a major new release of PopPUNK, which uses a new 'backend' to do sketching and distance calculation, [pp-sketchlib](https://github.com/johnlees/pp-sketchlib)

This changes the input format and the API somewhat, will be incompatible with previous database versions, and generates slightly different distance results. If you need backwards compatibility the previous version can still be run by specifying `--use-mash`.

New features:
* Use [pp-sketchlib](https://github.com/johnlees/pp-sketchlib) as the backend. This is ~2x faster for sketching and 50-100x faster for distance calculations. Databases are ~1/4 of the size.
* Input data is now formatted as a tab separated file with name followed by any associated sequence. Sample names no longer have to be filenames. (Closes 43, 46)
* Read data can now be handled, including a filter to remove k-mers containing sequencing errors
* Faster database edits with `prune_db` and `reference_pick`
* Ability to use the previous sketching method and databases by specifying `--use-mash`

Bug fixes:
* Better error handling when creating visualisations, so output files are still produced if this fails

This adds:
* The ability to use a simple core threshold (49)
* Use external clusters in generate viz mode (55)

Other:
* Add scripts into package (56)

The main reason for this release is the fix in 50. Although not necessarily a problem, in some populations the reference picking step can lead the pruned network to appear inconsistent with the clusters, giving warning messages when assigning new queries. This is fixed here.

Bug fixes:
* Fixes to `--generate-viz` mode (44).
* Problems using relative paths (48).
* Ensuring consistent network and cluster definitions when choosing references (50).

Other:
* Update travis-CI to use correct `DISPLAY` for `matplotlib`
* Add Azure pipelines CI