v0.1.4

Added

• Added cloud-streamable UK Biobank LD matrix paths hosted on Hugging Face, allowing users to pass
  hf://datasets/shz9/ukb-ld/<POP>/chr_*.zip directly to viprs_fit without pre-downloading archives.
• Added uv and Apptainer installation examples to the documentation.
• Added a DockerHub publishing workflow and modernized the CLI Docker image for release as
  shadizabad/viprs.
• Added uv.lock for reproducible development environments.
• Added tests to ensure runtime package version metadata stays synchronized.

Changed

• Restricted supported Python versions to Python 3.10 through 3.13, inclusive.
• Centralized package version metadata in viprs/_version.py and reused it across runtime and packaging metadata.
• Updated pyproject.toml metadata and dependencies to match the package's runtime, optional, test, and docs
  requirements.
• Split wheel build/testing and PyPI publishing into separate GitHub Actions workflows.
• Updated GitHub Actions runners and wheel builds to use current Linux, Windows, Intel macOS, and Apple Silicon
  macOS targets.
• Modernized the CLI Dockerfile to use python:3.11-slim-bookworm, OCI labels, configurable install targets,
  and installation smoke tests.
• Updated LD matrix documentation to put provenance and QC information before download/streaming instructions.

Fixed

• Fixed the binary f1 metric so it fits a logistic model on top of PRS values and thresholds predicted
  probabilities, instead of passing continuous PRS values directly to sklearn.metrics.f1_score.
• Made BLAS discovery in setup.py tolerate missing system pkg-config and fall back to the existing
  no-BLAS build path.