Optimize audio buffer handling to avoid repeated copies

[wpflueger]

Summary

• Replaced Buffer.concat([buffer, chunk]) on every PortAudio data event (O(n^2) worst-case) with a remainder-based approach
• The new code only concatenates the small leftover bytes (always < 1 frame = 640 bytes) with incoming chunks, not the entire accumulated buffer
• Functionally identical behavior — same frame extraction, same flush-on-stop logic

Test plan

• Record a dictation and verify audio frames are sent correctly (transcription works)
• Verify the remaining partial frame at end-of-stream is still flushed with silence padding
• No regressions in audio quality or timing

Closes #35

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Optimizes Electron audio capture buffering in createAudioStream to avoid repeated full-buffer concatenations on each PortAudio data event, reducing worst-case time complexity and copy overhead while preserving frame extraction and stop-flush behavior.

Changes:

• Replaced “grow-the-entire-buffer” concatenation with a small remainder-based buffering approach.
• Extracts frames via an offset into the current chunk (plus small remainder) and stores only leftover bytes for the next event.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The loop will never terminate if frameBytes is 0 (e.g., options.frameMs <= 0 or options.channels <= 0), because offset never advances. Please add input validation (throw early) or guard here to ensure frameBytes > 0 before entering the while loop.

[Copilot]

remainder is assigned as a subarray view into data. When data is chunk (or a large concat), this keeps the entire underlying buffer alive until the next data event even though only < frameBytes are needed. Consider copying the leftover bytes into a new small Buffer (bounded to < frameBytes) to avoid retaining potentially large backing buffers and to avoid relying on the lifetime/mutability of the native chunk buffer across async boundaries.

Suggested change

	// Keep only the leftover bytes (always < frameBytes)
	remainder = offset < data.length ? data.subarray(offset) : Buffer.alloc(0);
	// Keep only the leftover bytes (always < frameBytes), copying them into a new Buffer
	if (offset < data.length) {
	const remaining = data.length - offset;
	const newRemainder = Buffer.alloc(remaining);
	data.copy(newRemainder, 0, offset, data.length);
	remainder = newRemainder;
	} else {
	remainder = Buffer.alloc(0);
	}