v0.1.1 — convert streaming, docs/snapshot fallback

Added

• README badges (CI status, latest release, license, citation).

Changed

• Convert stage now streams parquet batches via pf.iter_batches() → RecordBatchReader → vxio.write instead of materialising whole tables.
  Resolves ArrowNotImplementedError: Nested data conversions not implemented for chunked array outputs from pyarrow on slugs whose nested columns
  (list<struct>, struct<bytes,…>) would need to be chunked across multiple
  Arrow arrays. Re-enables Vortex output for osm-germany-ways,
  ultrachat-200k, mmmu, websight-v01, peoples-speech-clean-validation.
• code-contests Vortex skip re-diagnosed: not the chunked-array path; a
  separate upstream FSST i32-offset overflow on list<string> >2 GB.
• open-food-facts description aligned with shipped output (currently a
  single raw_json: string column via jsonl_as_string_parse; VARIANT
  promotion deferred).
• PR template: dropped the "Test plan" checklist (CI runs the same gates on
  every PR; CONTRIBUTING.md documents them once).
• Agent-tooling docs (AGENTS.md, SKILLS.md, raincloud-docs skill) now flag
  docs/snapshot.json as load-bearing — TUI fallback and the
  row-count / file-size fallback for datasets.md regen. Stale "six derived
  docs" reference in AGENTS.md cleaned up to three.

Fixed

• docs/datasets.md regeneration now falls back to docs/snapshot.json
  (top-level scratch, then docs/v{schema_version}/snapshot.json on a fresh
  clone) for slugs whose parquet isn't built locally. Previously,
  partial-build regen would silently dash-out row counts and file sizes for
  any slug not on disk, destroying ground truth in the v1 snapshot. Snapshot
  regen now also captures last_built_row_groups. Five regression tests
  added in tests/test_docs.py.

v0.1.0 — initial public release

Initial public release.

Raincloud is a client-reproducible pipeline for building a curated catalog
of public datasets as analytics-ready Parquet + Vortex files. See
README.md for the user-facing overview,
AGENTS.md for the architecture, and
SKILLS.md for procedural playbooks.

This release bundles:

• The 7-stage build pipeline (fetch → extract → parse → transform → write
  → validate → convert) plus the optional opt-in hydrate stage.
• 249 dataset specs across 5 families (direct, kaggle-upstream,
  nyc-tlc, public-bi, uci).
• 24 named transform handlers covering CSV / Parquet / JSONL / XML / PBF /
  custom-format upstreams plus streaming variants for memory-constrained
  shapes.
• A read-only Textual TUI for browsing the catalog
  (python -m scripts.pipeline.browse, requires --extra tui).
• Per-dataset Vortex conversion via the convert.vortex flag.
• Apache License 2.0, with SPDX file headers on all Python sources.
• Governance: SECURITY.md, CONTRIBUTING.md, CODE_OF_CONDUCT.md
  (Contributor Covenant 2.1), DISCLAIMER.md (AS IS posture, content
  and license disclaimers, dataset-removal reporting), and
  HYDRATING.md (policy for the optional hydrate stage).
• Tooling: ruff lint (rules E, F, W, I) + GitHub Actions CI
  (.github/workflows/ci.yml) running lint, manifest validation, and
  pytest on every push and PR to develop.
• Dataset-removal issue template
  (.github/ISSUE_TEMPLATE/dataset-removal.yml) — structured form for
  the channel DISCLAIMER.md points readers at.
• Pull-request template (.github/pull_request_template.md) prompting
  for summary, test-plan checkbox list against the standard pre-PR gate,
  and change-type tags.
• CITATION.cff — GitHub-native citation metadata; surfaces the "Cite
  this repository" button in the repo sidebar with BibTeX / APA / Chicago
  exports.