csv-parser: add _detect_dialect helper

Heuristic refined from the plan draft: candidate header rows must be
followed by a row containing at least one numeric token. Without this,
IBKR-style multi-line preambles (all-text rows before the real header)
would be mistaken for the header at preamble=0.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

tests/test_llm_csv_parser.py

@@ -54,3 +54,45 @@ def test_fingerprint_is_sha256_hex_64_chars():
     f = _fingerprint(["Symbol", "Quantity"])
     assert len(f) == 64
     assert all(c in "0123456789abcdef" for c in f)
+
+
+def test_detect_dialect_no_preamble_comma():
+    from app.services.llm_csv_parser import _detect_dialect
+
+    raw = b"Symbol,Quantity,Avg Price\nAAPL,100,150.25\nMSFT,50,310.00\n"
+    delimiter, preamble = _detect_dialect(raw)
+    assert delimiter == ","
+    assert preamble == 0
+
+
+def test_detect_dialect_with_preamble():
+    from app.services.llm_csv_parser import _detect_dialect
+
+    raw = (
+        b"Statement,Header,Field Name,Field Value\n"
+        b"Statement,Data,BrokerName,Interactive Brokers LLC\n"
+        b"Statement,Data,Title,Activity Statement\n"
+        b"Statement,Data,Period,\"January 1, 2026 - January 31, 2026\"\n"
+        b"Symbol,Quantity,Avg Price,Currency,Description\n"
+        b"AAPL,100,150.25,USD,Apple Inc\n"
+    )
+    delimiter, preamble = _detect_dialect(raw)
+    assert delimiter == ","
+    # The data-row header line is the FIFTH line (index 4); preamble = 4.
+    assert preamble == 4
+
+
+def test_detect_dialect_tab_delimited():
+    from app.services.llm_csv_parser import _detect_dialect
+
+    raw = b"Symbol\tQuantity\tAvg Price\nAAPL\t100\t150.25\n"
+    delimiter, preamble = _detect_dialect(raw)
+    assert delimiter == "\t"
+    assert preamble == 0
+
+
+def test_detect_dialect_empty_raises():
+    from app.services.llm_csv_parser import LLMParseError, _detect_dialect
+
+    with pytest.raises(LLMParseError):
+        _detect_dialect(b"")