Reject out-of-range column index in Dataset.add_formatter

[uttam12331]

Summary

Dataset.add_formatter accepts a column index one past the last column, then crashes later on export instead of rejecting it up front.

Columns are 0-indexed everywhere (get_col(index) → row[index], formatter application → row[col] = handler(row[col])), so the maximum valid index is width - 1. But the bounds check uses <=:

if col is None or col <= self.width:      # accepts col == width (out of range)
    self._formatters.append((col, handler))
else:
    raise InvalidDatasetIndex

The else: raise InvalidDatasetIndex branch shows the intent is to reject out-of-range indices — the boundary is just off by one.

Reproduction

import tablib
d = tablib.Dataset()
d.headers = ['a', 'b']     # width 2 -> valid indices 0, 1
d.append([1, 2])
d.add_formatter(2, str)    # index 2 is out of range -> silently returns True
d.export('csv')            # IndexError: list index out of range (deep in format_row)

Expected: add_formatter(2, ...) should raise InvalidDatasetIndex at the call site.

Fix

-        if col is None or col <= self.width:
+        if col is None or col < self.width:

Tests

Added test_add_formatter_out_of_range_column: on a width-3 dataset it confirms the last valid index (2) is accepted and index 3 raises InvalidDatasetIndex. It fails on the current code (silently accepted) and passes with the fix; the existing formatter tests still pass.

[MohammedAlkindi]

#655 is the same fix, opened six weeks earlier (2026-07-10). Identical hunk in src/tablib/core.py: col <= self.width becomes col < self.width, +1/-1 on both. Whichever merges first will conflict the other.

The difference worth keeping is in this PR's test. It asserts width == 3, accepts add_formatter(2, ...) and rejects add_formatter(3, ...), so it pins both sides of the off-by-one. #655's test asserts only the rejection, and it also adds an AUTHORS entry.