Add --flush option to reduce stdout buffering latency

[phritz]

(AI-generated description, natch, but I think it covers it well. My use case is AEC and the ~50ms delay causes alignment headaches between reference audio from audiotee and what the mic is picking up)

PR Title

Add --flush option to reduce stdout buffering latency

PR Description

Summary

• Adds a --flush CLI flag that flushes stdout after each audio chunk write
• Fixes latency issues when piping AudioTee output to a parent process

Problem

When stdout is connected to a pipe (common when spawning AudioTee from Node.js, Python, etc.), the OS uses block buffering (typically 4KB-64KB). Since AudioTee writes 960-byte chunks (10ms at 48kHz s16le mono), 4-7 chunks accumulate before the buffer flushes, causing 40-70ms bursts instead of steady 10ms delivery.

This is problematic for real-time applications like:

• AEC (Acoustic Echo Cancellation) that rely on consistent timing between reference and capture signals
• Real-time ASR where latency spikes affect transcription timing
• Audio visualization where smooth updates are expected

Solution

Add fflush(stdout) after each FileHandle.standardOutput.write() call when the --flush flag is enabled.

Changes

• Sources/Output/Handlers/BinaryOutputHandler.swift: Add flushAfterWrite parameter to init, call fflush(stdout) when enabled
• Sources/CLI/AudioTee.swift: Add --flush flag, pass to BinaryAudioOutputHandler

Usage

# Enable flushing for real-time applications
./audiotee --flush | your_audio_processor

# Combine with other options
./audiotee --sample-rate 48000 --flush | node audio-processor.js

Tradeoffs

• (+) Enables real-time chunk delivery when piping
• (+) Backward compatible - default behavior unchanged
• (-) Slightly higher CPU overhead due to syscalls on each chunk (negligible for most use cases)

Testing

Tested by spawning AudioTee from Node.js and measuring inter-chunk timing:

• Without --flush: 40-70ms gaps between chunk arrivals
• With --flush: Consistent ~10ms gaps matching chunk duration