fix: suffixes_prefixes_titles always reflects current set state

[gaoflow]

Problem

Constants.suffixes_prefixes_titles cached its result in _pst after the
first access and never invalidated that cache. Any subsequent add() or
remove() call on titles, prefixes, suffix_acronyms, or
suffix_not_acronyms was silently ignored by the stale cache.

This creates an observable inconsistency: is_title() / is_prefix() /
is_suffix() query the live sets and return the correct answer, but
is_rootname() — which delegates to suffixes_prefixes_titles — keeps
returning the stale cached answer, causing it to contradict the other helpers.

Minimal reproduction:

from nameparser import HumanName
from nameparser.config import Constants

C = Constants()
# Access the property to prime the cache.
_ = C.suffixes_prefixes_titles

C.prefixes.add('xpfx')
hn = HumanName("", C)

# These two should agree — before this fix they don't:
print(hn.is_prefix('xpfx'))   # True  (correct — queries live set)
print(hn.is_rootname('xpfx')) # True  (wrong  — reads stale cache, should be False)

The same inconsistency affects titles and suffixes added or removed after the
property is first read. join_on_conjunctions relies on is_rootname to
count rootname_pieces and choose whether to join single-letter conjunctions,
so a stale _pst can silently skew that decision.

Fix

Drop _pst entirely and compute the set union fresh on every access. The
union of four SetManager instances is a simple O(n) operation with a small
constant and is called only during name parsing, so there is no meaningful
performance cost.

Tests

Five new tests in tests/test_constants.py verify that:

• suffixes_prefixes_titles reflects titles and prefixes added after
  construction.
• suffixes_prefixes_titles no longer contains a title that was removed.
• is_rootname and is_title/is_prefix remain consistent after
  add()/remove() calls.

All 682 tests pass (672 pre-existing + 10 from the dual-run fixture applied
to the 5 new test methods).

---

This pull request was prepared with the assistance of AI, under my direction and review.