v0.2.0

What's Changed

• Multi-GPU orchestration now auto-tunes max_in_flight based on the number
  of detected devices and keeps Dask clusters alive until every future is
  drained. This prevents premature shutdowns on longer studies and improves
  throughput on 4+ GPU systems.
• Added cohort-wide helpers in :mod:genboostgpu.tuning, including
  :func:~genboostgpu.tuning.select_tuning_windows for stratified sampling
  and :func:~genboostgpu.tuning.global_tune_params for Optuna-backed ridge
  refits derived from sparsity targets.
• Documentation now covers the reproducibility checklist, deterministic
  Optuna configuration, and richer tutorials linked from examples/ so new
  users can mirror the exact benchmarking pipelines.

Documentation Added

• Add Sphinx API documentation for backend by @KrotosBenjamin in #1
• Mock GPU dependencies for Sphinx autodoc by @KrotosBenjamin in #2
• docs: add quickstart and restructure guides by @KrotosBenjamin in #3
• docs: add quickstart and restructure guides by @KrotosBenjamin in #4

Full Changelog: v0.1.1...v0.2.0