Add opt-in acoustic echo cancellation to Swift WakeWordListener

[rajan9519]

What

Adds an opt-in echoCancellation flag to the Swift WakeWordListener.

Why

WakeWordListener captures raw microphone audio, so anything the device is
playing out — e.g. an assistant's own TTS — bleeds into the detection window
and can trigger false positives.

How

• New echoCancellation: Bool = false init parameter. When true, the mic is
  routed through the platform's voice-processing I/O unit
  (setVoiceProcessingEnabled(true)), which references the device output and
  subtracts it from the captured signal.
• Enabled before reading the input format, since toggling voice processing
  changes the node's format.
• On iOS the session mode switches from .measurement to .voiceChat, because
  .measurement disables system signal processing including AEC.
• If the platform can't enable it, start() throws the new
  WakeWordError.echoCancellationUnavailable rather than silently capturing raw
  audio, so callers that opted in aren't misled.
• README documents the flag.

Compatibility

Default is false, so existing callers are unchanged. No public API removed.

Testing

swift build and swift test pass (7 tests, 3 pre-existing skips). The
Python-only CI is unaffected.

[CLAassistant]

All committers have signed the CLA.