Add pyproject.toml packaging and pytest suite

- pyproject.toml: setuptools build, deps (requests), dev extra (pytest),
  pytest config with opt-in `live` marker
- tests/: 95 offline tests locking every series-ID encoding against
  known-good IDs (regression guard for the JOLTS/OES/ECI/productivity/QCEW
  fixes), plus a live API smoke test behind `-m live`
- README: install-as-package + test instructions; drop the now-resolved
  "no packaging"/"no tests" limitations
- gitignore build/test artifacts

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_live_smoke.py

@@ -0,0 +1,74 @@
+"""Live smoke test — hits the real BLS API.
+
+Opt-in only: deselected by default (see addopts in pyproject.toml). Run with:
+    pytest -m live
+Requires network and a BLS API key (BLS_API_KEY env var, or config.py).
+
+Guards against silent series-ID drift: BLS occasionally retires codes (e.g. the
+2010->2018 SOC change), which unit tests alone can't catch.
+"""
+import os
+
+import pytest
+
+from bls_client import BLSClient
+from bls_client.queries import employment as E, prices as P, wages as W, productivity as PR
+
+
+def _api_key():
+    key = os.environ.get("BLS_API_KEY")
+    if key:
+        return key
+    try:
+        from config import BLS_API_KEY  # local, gitignored
+        if BLS_API_KEY and BLS_API_KEY != "YOUR_KEY_HERE":
+            return BLS_API_KEY
+    except Exception:
+        pass
+    return None
+
+
+pytestmark = pytest.mark.live
+
+
+@pytest.fixture(scope="module")
+def client():
+    key = _api_key()
+    if not key:
+        pytest.skip("no BLS_API_KEY available")
+    return BLSClient(key)
+
+
+# One representative, previously-broken helper per repaired survey.
+REPAIRED = {
+    "JOLTS":        E.jolts_dashboard(),                       # dict of 5
+    "OES wage":     W.occupation_annual_median_wage("151252"),
+    "OES employ":   W.all_occupations_employment(),
+    "ECI":          W.eci_total_compensation(),
+    "Productivity": PR.business_output_per_hour(),
+    "QCEW":         E.qcew_national_private(),
+    "PPI final":    P.ppi_final_demand(),
+    "ECEC":         W.ecec_total_benefits(),
+}
+
+
+def _flatten(v):
+    return list(v.values()) if isinstance(v, dict) else [v]
+
+
+def test_repaired_surveys_return_data(client):
+    """Every repaired survey must return at least one real observation."""
+    all_ids = []
+    for ids in REPAIRED.values():
+        all_ids.extend(_flatten(ids))
+    res = client.fetch(all_ids, 2024, 2025)
+    missing = [sid for sid in all_ids if not res.get(sid, {}).get("data")]
+    assert not missing, f"no data returned for: {missing}"
+
+
+def test_headline_payrolls_plausible(client):
+    res = client.fetch_latest(E.nonfarm_payrolls(), years=1)
+    obs = client.latest_obs(next(iter(res.values())))
+    assert obs is not None
+    # US nonfarm payrolls are ~150,000 (thousands); sanity bound, not exact.
+    assert 100_000 < float(obs["value"]) < 200_000