fix(qwen-tts): serialize CUDA 13 flash-attn build

Limit the CUDA 13 FlashAttention source build to one ninja worker by default so it stays within CI runner memory. Preserve explicit MAX_JOBS overrides and leave other build profiles unchanged.

Assisted-by: Codex:gpt-5 [web]

backend/python/qwen-tts/install.sh

@@ -10,4 +10,10 @@ else
     source $backend_dir/../common/libbackend.sh
 fi
 
+# CUDA 13 has no prebuilt FlashAttention wheel, so the fallback source build
+# exceeds the CI runner's memory when ninja compiles multiple units at once.
+if [ "x${BUILD_PROFILE}" = "xcublas13" ]; then
+    export MAX_JOBS="${MAX_JOBS:-1}"
+fi
+
 installRequirements