Release 0.9.0

What's Changed

• data: move 74 public image/video datasets to legacy configs by @kenobijon in #131
• data: v23 3-class image / 4-class video taxonomy with audited labels by @kenobijon in #130
• data: add released v22 holdout datasets to public registry by @kenobijon in #129
• Clean up brittle and redundant tests by @dylanuys in #132
• Limit Hugging Face predownload request pressure by @dylanuys in #133
• Consolidate redundant Hugging Face dataset mirrors by @dylanuys in #134
• fix(data): remove vcc2020-real — 71% of it is converted speech by @kenobijon in #137
• feat(scoring): multiclass sn34 via Gorodkin MCC + multiclass Brier by @kenobijon in #136
• Release experimental taxonomy, scoring, and download updates by @dylanuys in #135

Full Changelog: v0.8.3...v0.9.0