Add a test to verify that clReleaseCommandQueue does trigger a flush in multi-queue scenario

[kpet]

See KhronosGroup/OpenCL-Docs#1248 for background.

We should add a test that:

• Creates two command queues
• Enqueues work in command queue A
• Enqueues work in command queue B that depends on work in command queue A
• Calls clReleaseCommandQueue on command queue A
• Calls clFinish on command queue B
• Checks that all work enqueued to command queue A and command queue B