fix(csv): handle trailing empty field at EOF without newline in COPY FROM

[chiangchenghsin-hash]

Summary

Fixes a CSV COPY FROM parser bug where a trailing empty field at end-of-file (without newline) is silently dropped, causing expected N values per row, but got N-1.

Bug

File: src/processor/operator/persistent/reader/csv/base_csv_reader.cpp — parseCSV() final_state

When a CSV file ends with a row like:

id,name,ver
1,a,
2,b,      ← no trailing newline, last field empty

The parser reaches final_state with position == start (both at EOF). The existing code only adds a remaining value when position > start, so the empty trailing field is skipped → column undercounts by 1 → addRow() throws "expected 3 values per row, but got 2".

Workaround users currently need

Quote the empty field as "" — this makes position > start true (parser moves past the closing quote), so the value is added. But this is fragile and not obvious.

Fix

Add an else if (column > 0) branch in final_state to emit an empty value when the file ends right after a delimiter:

     final_state:
         lineContext.setEndOfLine(getFileOffset());
         if (position > start) {
             // Add remaining value to chunk.
             if (!addValue(driver, curRowIdx, column,
                     std::string_view(buffer.get() + start, position - start - hasQuotes),
                     escapePositions)) {
                 return {curRowIdx, numErrors};
             }
             column++;
+        } else if (column > 0) {
+            // File ends right after a delimiter with an empty trailing field.
+            // Without this, a row like "a,b," (no trailing newline) loses its last
+            // empty field and undercounts columns, causing "expected N, got N-1".
+            if (!addValue(driver, curRowIdx, column, std::string_view{}, escapePositions)) {
+                return {curRowIdx, numErrors};
+            }
+            column++;
         }
         if (column > 0) {
             curRowIdx += driver.addRow(curRowIdx, column,
                 getOptionalWarningData<Driver>(columnInfo, option, getWarningSourceData()));
         }

Why this is safe

The else if (column > 0) branch only fires when all three are true:

1. position == start (file ended with no partial value to add)
2. column > 0 (we're mid-row, i.e. just past a delimiter — not at a clean row boundary)
3. We reached final_state (EOF)

This precisely matches "file ends right after a delimiter with an empty trailing field" and nothing else. Rows ending with \n are handled by add_row (which resets column to 0), so they don't hit this branch.

Verification

CREATE NODE TABLE t (id INT64, name STRING, ver STRING, PRIMARY KEY(id));
-- CSV content: "id,name,ver\n1,a,\n2,b,"  (no trailing newline)
COPY t FROM 'test.csv' (HEADER=true);
MATCH (n:t) RETURN count(n);  -- Before: error "expected 3, got 2". After: 2 rows copied ✅

Tested with 4 variants — only the no-trailing-newline case was broken, now all pass:

CSV variant	Before	After
LF + trailing newline	✅	✅
CRLF + trailing newline	✅	✅
Quoted empty ""	✅	✅
No trailing newline + empty last field	❌ expected 3, got 2	✅ 2 rows

Platform impact

Platform-agnostic. Affects all CSV COPY FROM users whose source files lack a trailing newline on the last row — a common situation when CSVs are generated programmatically (many emitters don't add a final \n).

[adsharma]

The windows style newlines in the code makes the diff hard to review. I ran dos2unix and added a test case in #648

[adsharma]

Thank you for your contribution!