v0.2.0 - new model and general improvements

What's Changed

• Add the Kumo adapter by @not-a-feature in #6
• Propagate model registry errors by @not-a-feature in #9
• Handle null dataset lists by @not-a-feature in #5
• Resolve registry exclusions by @not-a-feature in #7
• Fail early without AutoGluon by @not-a-feature in #8
• Preserve probability class labels by @not-a-feature in #11
• Measure the visible GPU by @not-a-feature in #12
• Use explicit regression groups by @not-a-feature in #10
• Configure training prevalence by dataset by @not-a-feature in #13
• Fingerprint dataset caches by @not-a-feature in #14
• Centralize model metadata by @not-a-feature in #15
• Generate web skill copies by @not-a-feature in #17
• Add benchmark doctor checks by @not-a-feature in #16
• Simplify artifact I/O by @not-a-feature in #19
• Use one fixed CV partition by @not-a-feature in #23
• Remove per-model training subsampling by @not-a-feature in #22
• Remove unused helpers by @not-a-feature in #20
• Use the model registry directly by @not-a-feature in #21

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Full Changelog: https://github.com/not-a-feature/TabBench-Bio/commits/v0.1.0