Fix 27s call ringing delay: TURN credential mismatch and STUN DNS TTL=0

Two bugs caused ICE candidate gathering to fail and timeout (~27 seconds)
before falling back to direct media on every call:

1. TURN credential mismatch — when TURN_PASSWORD was empty in .env,
   coturn defaulted to "changeme" but the entrypoint auto-generated a
   different random password for Asterisk. Every TURN auth attempt failed
   ("check_stun_auth: user easyasterisk credentials are incorrect").

2. STUN DNS TTL=0 — using the FQDN as stunaddr caused DNS resolution
   that returned TTL=0, making Asterisk cancel recurring STUN resolution
   entirely. Since coturn runs on the same host (network_mode: host),
   rtp.conf now uses 127.0.0.1 which needs no DNS at all.

Also documents the Android Call Integration audio issue (ConnectionService
routes audio through the native telephony path, breaking VoIP RTP).

https://claude.ai/code/session_01KWVtEt9MmZdywcu7WmgchX

docker-compose.yml

@@ -55,7 +55,7 @@ services:
       - TURN_ENABLED=y
       - TURN_SERVER=${DOMAIN_NAME:?}:${TURN_PORT:-3478}
       - TURN_USERNAME=${TURN_USERNAME:-easyasterisk}
-      - TURN_PASSWORD=${TURN_PASSWORD:-}
+      - TURN_PASSWORD=${TURN_PASSWORD:-changeme}
 
       # ── RTP Port Range ──
       - RTP_START=${RTP_START:-10000}