v0.9.0

Bugfix and training refactor release

• Fix readlink issue on MacOS
• Fix bug where medaka_variant did not call indels by default
• Fix bug in determining when to split contigs
• Drop support for older basecaller models (guppy<3.0.3)
• Store models in git-lfs
• Simplify medaka_variant workflow for speed
• Make network feature generation 2x faster
• Add smolecule command
• Log use of GPU and cuDNN, noting workaround for RTX cards
• Refactor labelling of training data and storing of models
• Reimplement RLE feature generation

Please do not attempt to use the "Source code" assets attached below for running medaka, they do not include medaka's model files.