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

[c-tonneslan]

Dataset.add_formatter guards the column index with col <= self.width, but column indices are 0-based, so the only valid indices are 0 .. width - 1. The <= lets col == width — always out of range — slip past the InvalidDatasetIndex guard.

The bad index isn't caught at registration; it surfaces later as a raw IndexError from row[col] when the formatter runs during export:

import tablib

d = tablib.Dataset()
d.headers = ['a', 'b']          # width 2, valid indices 0 and 1
d.append([1, 2])
d.add_formatter(2, lambda v: v * 10)   # index 2 == width, out of range
d.dict                                  # -> IndexError: list index out of range

Changing the guard to col < self.width rejects the invalid index at registration time with InvalidDatasetIndex, as intended. Valid indices, None (all columns), and header-name lookups are unaffected. InvalidDatasetIndex subclasses IndexError, so any caller already catching IndexError still works.

Added a regression test and myself to AUTHORS.