The a format token returns the literal string a instead of a lowercase meridiem indicator (am/pm).

>>> import pendulum
>>> dt = pendulum.datetime(2016, 8, 28, 23)
>>> dt.format("A")
'PM'
>>> dt.format("a")   # expected 'pm'
'a'

_format_localizable_token only handles the uppercase A token, so a falls through to the default branch that returns the token unchanged. The a token is documented and registered in the token regex, and the parser side already treats it as a lowercase meridiem, so formatting and parsing were inconsistent and a value formatted with a could not be round-tripped:

>>> pendulum.from_format(dt.format("YYYY-MM-DD hh:mm a"), "YYYY-MM-DD hh:mm a")

previously failed because the formatted string contained a literal a rather than pm.

This mirrors the existing A branch and lowercases the result for the a token. Added a regression test alongside the existing test_am_pm.