fix(tests): skip gitignored-dataset tests on CI (#47)

tests/benchmarks/datasets/ is gitignored, so CI never has the DBLP-ACM
files. Two tests (test_autoconfig_parity_pins_unchanged,
test_auto_configure_df_dblp_acm_does_not_crash) tried to read them and
failed with FileNotFoundError. Now they pytest.skip when the data file
is absent.

Both tests remain effective on developer machines that have pulled the
Leipzig dataset. Parity pin contract is unchanged — the capture script
at tests/parity/capture_autoconfig_output.py is still the source of
truth for regenerating pins.

Author: benzsevern

tests/test_autoconfig.py

@@ -1321,14 +1321,26 @@ def test_classify_by_data_prose_not_multi_name():
 def test_autoconfig_parity_pins_unchanged():
     """Pin test: AutoConfigDecisions refactor must not change output for the
     three benchmarks. If this fails, the refactor changed behavior - fix the
-    refactor, not the pin file."""
+    refactor, not the pin file.
+
+    Benchmark datasets are gitignored; this test is effectively local-dev
+    only. CI skips it gracefully. Contributors who want to exercise the
+    parity guard should run `python tests/parity/capture_autoconfig_output.py`
+    after pulling the Leipzig DBLP-ACM dataset into tests/benchmarks/datasets/.
+    """
     import json
     import sys
     from pathlib import Path
 
     import polars as pl
+    import pytest
 
     repo_root = Path(__file__).parent.parent
+    dblp_path = repo_root / "tests" / "benchmarks" / "datasets" / "DBLP-ACM" / "DBLP2.csv"
+    if not dblp_path.exists():
+        pytest.skip(
+            "DBLP-ACM dataset not present (gitignored, local-dev only)"
+        )
     # Import pin_config from the capture script (not a normal test import)
     sys.path.insert(0, str(repo_root / "tests" / "parity"))
     from capture_autoconfig_output import pin_config  # type: ignore

tests/test_autoconfig_verify.py

@@ -408,9 +408,18 @@ def test_preflight_check5_drops_empty_matchkey_after_record_embedding_removal():
 
 
 def test_auto_configure_df_dblp_acm_does_not_crash():
+    """Regression test for the v1.5.0 preflight fix: zero-config dedupe_df
+    on biblio-style data no longer crashes with 'Missing required columns'.
+
+    Benchmark datasets are gitignored; CI skips gracefully. Full integration
+    coverage lives in tests/test_autoconfig_benchmarks.py (marked @benchmark).
+    """
     from pathlib import Path
+    import pytest
     from goldenmatch._api import dedupe_df
     d = Path("tests/benchmarks/datasets/DBLP-ACM")
+    if not (d / "DBLP2.csv").exists():
+        pytest.skip("DBLP-ACM dataset not present (gitignored, local-dev only)")
     dblp = pl.read_csv(d / "DBLP2.csv", encoding="utf8-lossy", ignore_errors=True)
     acm  = pl.read_csv(d / "ACM.csv", encoding="utf8-lossy", ignore_errors=True)
     df = pl.concat([dblp, acm], how="diagonal_relaxed")