feat(asr): add pluggable ONNX streaming backend and model registry

Add a second backend (sherpa-onnx) alongside llama.cpp, so native
streaming models can be selected by ID from a JSON registry. All such
models share one adapter: only the model files and a few options differ.
Models now live under the app data directory and are fetched by
scripts/fetch-models.sh. Adding a model is a data change, not code.

Measured on the same audio: the streaming backend is ~35x cheaper in CPU
than Qwen3-ASR (0.10-0.12 vs 4.27 core-seconds per audio second), shows
the first caption at 0.48-1.44s, and matches its accuracy for zh/en.

新增 sherpa-onnx 流式后端，与 llama.cpp 并存；模型按注册表 id 从应用
数据目录加载。同族模型共用一个适配器，新增模型只改 data/models.json。
实测流式后端 CPU 约为 Qwen3-ASR 的 1/35，首个字幕 0.48~1.44s，中英
准确率持平。注意预编译库是旧 C++ ABI，必须走其 C API。

scripts/fetch-models.sh

@@ -0,0 +1,74 @@
+#!/usr/bin/env bash
+# 把模型注册表里的模型下载到应用数据目录（QStandardPaths::AppDataLocation）。
+#
+# 用法:
+#   ./fetch-models.sh                 # 列出所有模型和安装状态
+#   ./fetch-models.sh <id> [...]      # 下载指定模型
+#   ./fetch-models.sh --all           # 下载全部
+#
+# 目录规则（与程序里 ModelStore 保持一致）：
+#   ${XDG_DATA_HOME:-~/.local/share}/live-caption/models/<id>/
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+REG="$ROOT/data/models.json"
+DEST="${XDG_DATA_HOME:-$HOME/.local/share}/live-caption/models"
+
+[ -f "$REG" ] || { echo "找不到注册表: $REG" >&2; exit 1; }
+
+ids=$(python3 -c "
+import json,sys
+d=json.load(open('$REG'))
+print(' '.join(m['id'] for m in d['models']))")
+
+show_status() {
+  python3 - "$REG" "$DEST" <<'PY'
+import json, os, sys
+reg, dest = sys.argv[1], sys.argv[2]
+d = json.load(open(reg))
+for m in d['models']:
+    mid = m['id']; mdir = os.path.join(dest, mid)
+    files = m['files']
+    have = sum(1 for f in files.values() if os.path.exists(os.path.join(mdir, f['name'])))
+    total = len(files)
+    mark = "已安装" if have == total else ("部分" if have else "未安装")
+    langs = ",".join(m['languages'][:4]) + ("…" if len(m['languages']) > 4 else "")
+    print(f"  [{have}/{total} {mark}] {mid:22s} {langs:18s} {m.get('license','')}")
+    print(f"        {m['name']}")
+PY
+}
+
+download_one() {
+  local id="$1"
+  local dir="$DEST/$id"
+  mkdir -p "$dir"
+  echo "==> $id → $dir"
+  python3 - "$REG" "$id" <<'PY' | while IFS=$'\t' read -r name url; do
+import json, sys
+d = json.load(open(sys.argv[1])); mid = sys.argv[2]
+m = next(x for x in d['models'] if x['id'] == mid)
+for f in m['files'].values():
+    print(f"{f['name']}\t{f['url']}")
+PY
+    if [ -f "$dir/$name" ]; then
+      echo "    已存在，跳过: $name"
+      continue
+    fi
+    echo "    下载 $name"
+    if command -v aria2c >/dev/null 2>&1; then
+      aria2c -x16 -s16 -k4M --summary-interval=0 -d "$dir" -o "$name" "$url" >/dev/null
+    else
+      curl -fL --progress-bar -o "$dir/$name" "$url"
+    fi
+  done
+  echo "    完成"
+}
+
+case "${1:-}" in
+  "" ) echo "模型存储目录: $DEST"; echo; show_status ;;
+  --all) shift; for id in $ids; do download_one "$id"; done ;;
+  *) for id in "$@"; do
+       echo "$ids" | tr ' ' '\n' | grep -qx "$id" || { echo "未知模型: $id" >&2; exit 1; }
+       download_one "$id"
+     done ;;
+esac