feat: add optional Laya engine mode for grouping and language detection (#12)

Add a languageDetectionMode setting (algo | laya | laya+algo) exposed
in Settings with three engine choices. When Laya is enabled, a laya
service (Dockerfile.laya, optional compose profile, models stored
outside the repo under ./data/models/laya) can be started, with model
download and deletion from the UI via /laya routes to manage disk
space. Add a channelLanguageFusion service combining Laya results
with the existing algorithm, and wire detection mode through
grouping, channel list and player. Cover engine selection, fusion
and laya service behavior with tests.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Dockerfile.laya

@@ -0,0 +1,23 @@
+FROM python:3.12-slim
+
+ARG TORCH_VERSION=2.14.1
+
+# Install the CPU wheel explicitly before Laya so pip does not select the much
+# larger CUDA build. Model weights remain runtime-only in the mounted cache.
+RUN python -m pip install --no-cache-dir \
+        "torch==${TORCH_VERSION}" \
+        --index-url https://download.pytorch.org/whl/cpu \
+    && python -m pip install --no-cache-dir "laya[serve]>=0.3.22,<0.4" \
+    && python -m pip check \
+    && python -c "import torch; assert torch.version.cuda is None, torch.version.cuda"
+
+ENV HF_HOME=/models \
+    LAYA_HOST=0.0.0.0 \
+    LAYA_PORT=8000 \
+    LAYA_PRELOAD=0 \
+    LAYA_DEVICE=cpu
+
+VOLUME ["/models"]
+EXPOSE 8000
+
+CMD ["laya-serve"]