fix: normalize task hash in CUR cost JOIN by stripping '/'

Task hashes are Nextflow workdir paths ('ab/cdef12'), CUR hashes are
plain MD5 prefixes ('abcdef12'). REPLACE(t.hash, '/', '') before LEFT()
ensures the JOIN matches correctly.

Author: edmundmiller

bin/benchmark_report.py

@@ -328,7 +328,7 @@ def query_run_costs(db: duckdb.DuckDBPyConnection) -> list[dict]:
                 ROUND(SUM(COALESCE(c.unused_cost, 0)), 2) AS unused_cost,
             FROM runs r
             LEFT JOIN tasks t ON r.run_id = t.run_id
-            LEFT JOIN costs c ON t.run_id = c.run_id AND LEFT(t.hash, 8) = c.hash
+            LEFT JOIN costs c ON t.run_id = c.run_id AND LEFT(REPLACE(t.hash, '/', ''), 8) = c.hash
             GROUP BY r.run_id, r."group"
             ORDER BY r."group"
         """)

bin/test_benchmark_report.py

@@ -138,7 +138,6 @@ def _write_cur_parquet(self, tmp_path, run_id="run1", task_hash_md5="abcdef12345
         db.close()
         return path
 
-    @pytest.mark.xfail(strict=True, reason="CUR hash-join mismatch: LEFT(t.hash,8) includes '/' so never matches CUR hash")
     def test_cur_costs_override_task_costs(self, tmp_path):
         """When CUR data is present, query_run_costs should use CUR cost (10.0)
         not the task-level cost (1.50)."""
@@ -153,7 +152,6 @@ def test_cur_costs_override_task_costs(self, tmp_path):
         # CUR total cost = split_cost + unused = 8.0 + 2.0 = 10.0
         assert costs[0]["cost"] == pytest.approx(10.0)
 
-    @pytest.mark.xfail(strict=True, reason="CUR hash-join mismatch: LEFT(t.hash,8) includes '/' so never matches CUR hash")
     def test_cur_used_and_unused_cost_split(self, tmp_path):
         """CUR provides used_cost and unused_cost breakdown."""
         task = _flat_task(hash_val="ab/cdef1234567890", cost=1.50)