v0.1.0

First public release. Published to PyPI as microtrade-fwf; the import path
and CLI command stay microtrade.

Added

• Streaming FWF ingest pipeline (microtrade ingest) that reads each
  <trade_type>_<YYYYMM>.zip directly from its archive, slices columns
  against a versioned spec, and writes year=YYYY/month=MM/part-0.parquet
  atomically under a per-type dataset root. Per-partition JSONL manifest
  under <output>/_manifests/<trade_type>/<run_id>.jsonl.
• Excel → YAML spec converter (microtrade import-spec) for the real
  upstream workbook layout: positional sheet→trade_type mapping,
  autodetected Position | Description | Length | Type header, skipped
  Blank filler rows, Char/Num dtype tokens.
• Per-sheet preamble sanity check that refuses to accept a mis-ordered
  workbook (catches swapped imports/exports sheets before they land in
  YAML).
• Date dtype (pyarrow date32) with yyyymmdd_to_date /
  yyyymm_to_date parsers, resolved per column at stream start.
• Row-level quality-issue log: parse failures (bad numeric, blank
  non-nullable, bad date) go to
  <output>/_quality_issues/<trade_type>/<run_id>.jsonl and the row is
  skipped; the surrounding partition still writes. Capped per partition
  (--max-quality-issues, default 10,000) and guarded by a skip-rate
  abort (--max-skip-rate, default 0.5) so pathological inputs can't
  balloon the log or silently land a mostly-broken partition.
• Canonical _dataset_schema.json refreshed at
  <output>/<trade_type>/_dataset_schema.json after every run — union
  of all committed specs for that trade type, latest dtype wins,
  nullability widens.
• microtrade validate-specs — lints every v*.yaml under a spec dir,
  verifies filename versions match effective_from, prints a
  per-trade-type diff_specs changelog, and checks canonical_columns
  to surface cross-version dtype conflicts. Summary reports scan scope
  (OK (N trade types, M specs)).
• microtrade inspect — dumps the resolved spec and first rows of a raw
  zip (or plain FWF via --type / --period), with per-column
  [start..end] dtype 'value' annotation; --raw prints full lines.
• Reference YAML specs generated from examples/microdata-layout.xls
  ship under src/microtrade/specs/, so microtrade ingest works out
  of the box against any real data that matches the 2020-01 layout.

Notes

• Consumers that hive-scan the dataset should use a **/*.parquet glob
  because _dataset_schema.json lives at the dataset root (see the
  README's "Output layout" section).
• All four CLI subcommands (ingest, import-spec, validate-specs,
  inspect) ship implemented; the package is fully typed (py.typed
  marker included in the wheel).