v0.9.0

What's Changed

• docs: endor user-report needs ENDOR_API_KEY by @rasmus-u in #29
• feat: fine-tuning with images (endor.Image in rows and datums, uploaded once) by @rasmus-u in #30
• feat: project.learning() and no more pausing by @rasmus-u in #31
• Clear errors for a label in the wrong shape, and a warning for options with no training labels by @rasmus-u in #33
• decision_metrics: a tied top shares the credit, so accuracy doesn't depend on option order by @rasmus-u in #34
• client.corrections (replaces feedback), decision ids, learning drift, one base model per continuous learning project by @rasmus-u in #35
• 0.9.0: continuous learning projects with live and preview versions; custom models outside projects by @rasmus-u in #36
• fix: soak findings (serving state, continuous_learning, visible GPU waits, runs.get heartbeat, CLI tables) by @rasmus-u in #37
• BaseModelInfo.decision_index and tokens_per_s; endor base-models table; trainer_gpu gone by @rasmus-u in #38
• 0.9.0 check findings: corrections wait for their decision, -f after any command, ResponseValidationError everywhere, async across event loops; image training out of 0.9.0 by @rasmus-u in #40
• feat: base models by owner/model ids (mapika/decider-2b), called directly; ModelRequiresProjectError gone by @rasmus-u in #41
• feat: warn (DeprecationWarning) when a response carries a Deprecation header by @rasmus-u in #39
• feat: release 0.9.0 by @rasmus-u in #42

Full Changelog: v0.8.0...v0.9.0