I found that ー, ゠ and ・ are not detected with \p{Katakana} but as range \x{30A0}-\x{30FF}.
See https://regex101.com/r/PZzTLm/1 and http://www.rikai.com/library/kanjitables/kanji_codes.unicode.shtml
I can't find anything on this. Does anyone have a source that explains why these characters are not included? The problem is not unique to \p{Katakana}. \p{Hiragana} and others have similar issues.
In \p{Katakana}, \x{30A1}-\x{30FA}\x{30FD}-\x{30FF} is used instead of the \x{30A0}-\x{30FF} range, where \x{30A0}, \x{30FB} and \x{30FC} are excluded.
There is no reason these chars should not have been included because when you use \p{Block=Katakana} Katakana script block Unicode property class you will match all chars in the \x{30A0}-\x{30FF} range.
If you may actually combine the two, [\p{Katakana}\p{Block=Katakana}], you will match all the chars you expect.
If you use ECMAScript regex flavor, the implementation is
\p{scx=Katakana}
See the regex demo. The scx prefix means all indicated script extensions are included:
Scx set contains multiple explicit Script values; Script(cp) is implicit
and
For example, U+30FC KATAKANA-HIRAGANA PROLONGED SOUND MARK is shared across the Hiragana and Katakana scripts, but is not used in other scripts, so it is assigned an scx set value of {Hira Kana}.