v4.0.2 - Epic-direct artifact ids and stub-wording fix

Fixed

• consistency-check reads epic-direct fix and improvement ids. A fix
  does not always hang off a feature: projects that attach one directly
  to its epic write FIX-{ee}-{nn}, and the backlog scanner has always
  accepted that shape, but parse_artifact_id only knew
  FIX-{ee}-{ff}-{nn}. The two disagreed, so every epic-direct row was
  reported as an orphan even when its artifact file sat right next to
  the others, and no file could ever satisfy such a row. Both depths are
  accepted now, deeper pattern first so FIX-19-01-01 cannot degrade
  into FIX-19-01. The sub-feature form FEAT-{ee}-{ff}-{nn} is
  accepted alongside it, and renumber-for-merge scans the same two
  depths so its duplicate check no longer walks past epic-direct files.
• N-19 no longer reads a test double as unfinished work. The rule
  recognises a backlog row that documents a stub by its wording and
  matched a bare "stub" anywhere in the line, so a row saying
  "Regressionstest plus Modal-Stub in tests/stubs/obsidian.ts" demanded
  a FIXME(stub) marker for work that was already done. Bare "stub"
  now counts only as a standalone word.

Added

• tools/tests/test_id_patterns.py and tools/tests/test_stub_notes.py
  cover both id depths, the precedence between them, and the wording
  that separates unfinished work from test doubles.