fix(transmit): exclude FreeDV modes from Quindar tone allowlist

[NF0T]

Summary

Closes #3318.

isPhoneModeForQuindar() previously allowed Quindar intro/outro tones to fire in
FreeDV modes (FDV, FDVU, FDVL). FreeDV now uses RADAE — the same neural speech
autoencoder as RADE mode. The Quindar tone is injected into the TX audio stream
before it reaches the radio, where the radio's internal RADAE encoder processes it.
A 2525/2475 Hz sine fed into a neural encoder trained on speech produces
codec-artifact noise on air, not a recognisable signalling tone.

RADE mode was already correctly excluded (it uses DIGU/DIGL, never in the allowlist).
FreeDV was explicitly included, likely when FreeDV still used Codec2. This removes
the three FreeDV mode strings and updates the comment to explain the RADAE connection.

Impact

No user-visible Quindar behaviour was working correctly in FreeDV modes before this
change — the tone was being mangled by RADAE before transmission. This removes a
broken feature, not a working one.

SSB/AM/FM Quindar behaviour is unchanged.

Test plan

• SSB TX with Quindar enabled: intro/outro tones fire normally (no regression)
• FreeDV TX with Quindar enabled: no Quindar tones fire (expected new behaviour)

Refs

• Issue fix(transmit): exclude FreeDV modes from Quindar tone allowlist #3318
• PR fix(transmit): route Quindar outro timer through dispatchMoxOff() #3317 — dispatchMoxOff() refactor where this was surfaced
• TransmitModel.cpp::isPhoneModeForQuindar() — the only function changed

🤖 Generated with Claude Code

[jensenpat]

No findings.

I traced the Quindar gate from TransmitModel::requestPttOn()/requestPttOff() through the TX slice mode getter and the AudioEngine insertion point. Removing FDV/FDVU/FDVL from isPhoneModeForQuindar() means FreeDV no longer starts intro/outro phases, so the Quindar processor remains idle and the PTT path falls straight through to MOX. USB/LSB/AM/FM/NFM behavior is unchanged.

I also checked the RADE side: client RADE still activates on DIGU/DIGL and takes the early RADE TX branch, so it remains outside the Quindar voice-chain path. The current upstream main already contains #3317, and this PR applies as a single allowlist/comment change on top of it.

CI, macOS/Windows checks, and CodeQL are green on b3a1d2f. The only residual gap is that the private mode predicate still lacks direct unit coverage for the FDV false cases, but I do not think that blocks this one-line fix.