Currently, GpuMediaService is configured with |use_own_thread| = true, so that it is not blocked by (and conversely does not block) the GPU main thread.

However, since GpuMediaService also shuts itself down when inactive, auxiliary objects that outlive media services cannot rely on the SingleThreadTaskRunner sticking around. The result is passing service_manager::ServiceContextRef objects everywhere to prevent the thread from dying.

There are two obvious alternatives:
  1) Don't shut down the thread at all. Saving one thread may not be enough to justify the effort. We can still destroy all of the service objects on a timeout if desired.
  2) Switch to a SequenceTaskRunner running on a long-lived thread pool.

Since (2) appears to be the target state for Chromium overall, this bug tracks implementing that solution.