feat(apple): make SpokenInstructionObserver's audio session management opt-out

[meme-te]

Problem

SpokenInstructionObserver always takes over the shared AVAudioSession around each spoken
instruction. That is the right default for apps using the built-in AVSpeechSynthesizer, but it
permanently strands other apps' audio for apps that inject a custom SpeechSynthesizer playing
through their own session.

The mechanism:

• requestAudioFocus() sets .duckOthers + .interruptSpokenAudioAndMixWithOthers and
  mode = .voicePrompt, then activates the session.
• releaseAudioFocus() clears hasAudioFocus only after setActive(false) succeeds.
• setActive(false) fails while the session still has active audio I/O.

An app that keeps a microphone tap open (wake-word standby) or plays its own guidance audio
therefore never releases focus: hasAudioFocus stays true, every later requestAudioFocus()
early-returns on guard !hasAudioFocus, and other apps stay ducked for the rest of the session.

We measured this on a device with a Bluetooth headset: background music dropped from 0.80 to
0.30 system output volume about one second after navigation started, and never recovered.

Recovering from the host app does not work either. Re-applying setCategory on an already-active
session interrupts other apps' playback, so "take the session back" trades a stuck duck for
repeated audio dropouts (we tried it, and it was worse).

There is also no way to opt out today: FerrostarCore takes spokenInstructionObserver as a
non-optional concrete SpokenInstructionObserver, and the class is public but not open, so it
can be neither replaced nor subclassed from another module.

Change

Adds managesAudioSession to SpokenInstructionObserver, defaulting to true so existing
behavior is unchanged. Apps that own their audio session can pass false:

SpokenInstructionObserver(synthesizer: mySynthesizer, isMuted: false, managesAudioSession: false)

One file, guarding the three audioManager calls. No change to instruction de-duplication or any
other logic.

Testing

Built and run on device (iPhone, Bluetooth A2DP headset) with managesAudioSession: false:
background music keeps playing across app launch and repeated microphone on/off cycles, and the
app's own spoken guidance still plays through its session.