improves handling of command buffer emulation events

[bashbaug]

This PR improves handling for command buffer emulation events. Specifically:

1. The event type for a command buffer emulation event is now properly returned (CL_COMMAND_COMMAND_BUFFER_KHR).
2. Event profiling for the QUEUED, SUBMITTED, and START times is more likely to be accurate.

Note that (2) does require event profiling to be properly implemented for barrier events, and empirically this does not seem to be happening for all OpenCL implementations.