TensorFold 0.6.6: --name-priority on the CUDA server

• --name-priority ID=background on the CUDA server. A request that names that served id (--name or an
  --alias) and sends no priority of its own is served as priority: "background", so it yields to foreground
  requests. A request's own priority always wins. This suits clients that can choose a model id but cannot add a
  field to the request, such as batch extractors. On one DGX Spark, a foreground request behind four background ones
  got its first token in 0.2 s instead of 21 s, with replies unchanged (#445). Thanks to @philip-pentatonic.