test(const): the AppType split tests pin IANA member counts, so a vendor update turns them red #1820

tests/const/test_const_apptype_split_unit.py hardcodes IANA's per-registry member counts: {'TCP': 6147, 'UDP': 6143, 'SCTP': 91, 'DCCP': 10}, plus a 12,391 total and a 766 that is also count-derived (lines 868, 981-992, 1264, 1295, 1338, 1370). The values come from upstream data, not from the code under test, so any vendor update that adds or removes a port breaks the test even when the code is correct.

That has now happened. b0256f891 (the 1.5.0b13 bump) added port 4930 (TCP.fmsg, UDP.reserved_4930). On that commit, 5 tests in the file fail with 6148 != 6147 : TCP population changed, 12393 != 12391 and 767 != 766. I reproduced this locally on origin/main. Main's Unit Tests run 38874587976 for that commit is still in progress, so expect it to go red.

Fix: make the expectations count-free by deriving them from the registries themselves. Keep each test's original defect detection: #806 (composite proto), #807 (aliases), and the extra-member regression that the hardcoded len(TCP) caught. A mutation run must still fail. The docstrings repeat the same counts, and those go too.