Integrate YuE2 as an exclusive Athena mode

compose.yaml

@@ -487,10 +487,12 @@ services:
       RESTORE_WORKER: restore
       TTS_WORKER: qwen3
       MUSIC_WORKER: acestep
+      YUE2_WORKER: yue2
       SEPARATOR_WORKER: bs-roformer
       VOICE_WORKER: vevo2
       VOICE_CHANGE_WORKER: xvc
       APPLIO_WORKER: applio
+      TRELLIS_WORKER: trellis2-q8
     networks: [control]
     security_opt: ["no-new-privileges:true"]
     healthcheck:
@@ -526,6 +528,8 @@ services:
       PROFILE_CONTROL_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
       SWITCH_TIMEOUT: "600"
       REQUEST_TIMEOUT: "600"
+      YUE2_START_TIMEOUT: "600"
+      TRELLIS_START_TIMEOUT: "900"
       # Last-resort guard for every OpenAI-compatible client. Without a
       # request limit llama.cpp uses n_predict=-1 and a reasoning loop can
       # consume the complete context before yielding visible output.
@@ -748,6 +752,8 @@ services:
       VOICE_CHANGE_UI_URL: "${VOICE_CHANGE_UI_URL:-http://192.168.1.212:8009/}"
       APPLIO_UI_URL: "${APPLIO_UI_URL:-http://192.168.1.212:8011/}"
       MIKES_APPLIO_UI_URL: "${MIKES_APPLIO_UI_URL:-http://192.168.1.212:8012/}"
+      TRELLIS_UI_URL: "${TRELLIS_UI_URL:-http://192.168.1.212:8013/}"
+      YUE2_UI_URL: "${YUE2_UI_URL:-http://192.168.1.212:8014/}"
       HOST_PROC: /host/proc
       HOST_DATA: /host/data
       HOST_MODELS: /host/models