csvbase: report the correct line number in extraction errors

[MugenLab]

Importer.extract() computes lineno as the 1-based physical line number of
the data row:

offset = int(self.header) + bool(self.names) + 1
for lineno, row in enumerate(self.read(filepath), offset):

That is also the number metadata() records through
data.new_metadata(filepath, lineno), so it is what ends up in the metadata of
the extracted transactions. The error path, however, formatted lineno + 1:

raise RuntimeError(
    f"Error processing {filepath} line {lineno + 1} with values {row!r}"
) from ex

so the same row was reported at two different line numbers depending on whether
it parsed or not, and the error pointed one line past the offending row.

Reproduced with beancount 3.2.3 on a three-line file (header plus two data rows,
the last one invalid):

physical line 1: date,narration,amount
physical line 2: 2025-01-25,Good,-1.00
physical line 3: 2025-01-26,Bad,invalid

Error processing /tmp/t.csv line 4 with values ('2025-01-26', 'Bad', 'invalid')
metadata for the row on physical line 2 -> {'filename': '/tmp/t.csv', 'lineno': 2}

The existing test_report_exception encoded the old behaviour (its docfile has
two lines and the failing row is the second one, but the expected message said
line 3), so its expectation is updated to line 2.

Two regression tests are added:

• test_report_exception_matches_metadata_lineno extracts a file successfully,
  remembers the lineno metadata of the last row, then makes that same row
  invalid and requires the reported line number to be the one from the
  metadata. This pins the two code paths together rather than hard-coding a
  constant, so they cannot drift apart again.
• test_report_exception_line_numbering covers a file that has both skipped
  header lines (header = 1) and a names row, and asserts against the line
  index actually read back from the file.

Without the one-character fix all three tests fail; with it the full suite
(pytest beangulp examples, 174 tests) passes, and ruff check beangulp/ examples/ is clean.

---

Disclosure: this change was written by an AI agent. A human is accountable for
the submission and will respond to review comments.