Add WSL and Linux support to VoiceServer

[selfscrum]

Summary

• Add platform detection (macos, wsl, linux)
• WSL: Play audio via Windows PowerShell MediaPlayer (bypasses WSLg crackling)
• Linux: Play audio via mpv
• macOS: Keep existing afplay behavior
• Use PAI_DIR environment variable for all paths (with ~/.claude fallback)
• Load .env from both ~/.env and $PAI_DIR/.env
• Add platform info to /health endpoint

Test plan

• Tested on WSL2 - voice plays correctly through Windows MediaPlayer
• Test on macOS (existing behavior should be unchanged)
• Test on native Linux with mpv installed

🤖 Generated with Claude Code

[fayerman-source]

Great work on the Linux/WSL audio support! I tested a similar approach locally and can confirm mpv works well.

Observation: While testing on Ubuntu 24.04/WSL2, I noticed two additional macOS-specific dependencies that still cause errors on Linux:

1. Notification Display (osascript)

ENOENT: no such file or directory, posix_spawn '/usr/bin/osascript'
spawnargs: [ "-e", "display notification \"...\" with title \"PAI Notification\"" ]

Suggested fix: Use notify-send on Linux:

if (isMac) {
  spawn('/usr/bin/osascript', ['-e', `display notification "${message}" with title "${title}"`]);
} else {
  spawn('notify-send', [title, message]);
}

2. Fallback TTS (say)

The fallback speakWithSay() function uses /usr/bin/say which doesn't exist on Linux.

Suggested fix: Use espeak or spd-say on Linux:

if (isMac) {
  spawn('/usr/bin/say', [text]);
} else {
  spawn('espeak', [text]); // or 'spd-say'
}

These are non-blocking errors (the server continues to work), but addressing them would complete the cross-platform support. Happy to submit a follow-up PR if helpful!