…#3888). Principle VI.

Closes two pre-existing stuck-TX windows in the momentary keyboard-keying
family (PTT (Hold), CW straight key / paddles), surfaced by #3884's
rebindable PTT-hold key.

1. Focus loss / deactivation while held — the KeyRelease that would un-key
   goes to whatever app gains focus and never reaches our event filter, so
   the flag stayed set and TX stayed keyed. Add failSafeMomentaryKeyingToRx():
   on QEvent::ActivationChange (window deactivated) and, as a macOS
   belt-and-suspenders, QEvent::ApplicationStateChange (app backgrounded),
   clear every momentary flag and issue the matching un-key. It reuses the
   existing requestPttOff / setCw*State paths so the interlock gating and
   Quindar outro still run, and no-ops when nothing is keyed.

2. Modifier-combo binding released modifier-first — a binding like Ctrl+T
   released Ctrl-first delivers a KeyRelease whose QKeySequence(modifiers|key)
   is now plain T and no longer matches Ctrl+T, so the un-key never fired.
   While a momentary action is active, also accept a release matched by the
   bound base key alone (keyEventMatchesActionBaseKey), so releasing any part
   of the combo fails safe to RX.

Client-side only; the un-key terminates in the existing xmit / sendCwKey
paths with their interlock gating intact. A held transmit key must never
strand the transmitter — fail-safe-to-RX is the Principle VI expectation.

Blast radius: risk_score=0.044 on handlePttHoldShortcut, 3 high-risk
affected (top: MainWindow::eventFilter, RadioModel::transmitModel,
RadioModel::isConnected). All are the existing caller/un-key path this
change reuses; no signatures change and the new changeEvent branch is a
separate ActivationChange early-path that never touches the minimal-mode
AppSettings logic.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>