fix: handle empty and headerless datasets in DBF export

[santhreal]

DBF export_set inferred each field's type from dataset[0]. On a headers-only dataset (columns defined, no rows) that raised IndexError, and because the function never checked for headers, a headerless dataset raised a raw TypeError: 'NoneType' object is not iterable from zip(dataset.headers, ...).

Two changes: default field types to character when there are no rows (a headers-only dataset now exports to a valid dbf that round-trips), and raise HeadersNeeded when the dataset has no headers, since dbf fields must be named (matches the HeadersNeeded pattern used elsewhere rather than leaking a TypeError).

import tablib
d = tablib.Dataset(); d.headers = ['name', 'gpa']
d.export('dbf')  # IndexError before, valid dbf after
tablib.import_set('[[1,2]]', format='json').export('dbf')  # TypeError before, HeadersNeeded after