This updates the ROC target documentation to reflect the work on
dGPUs and retires the work on APU. It:

* Accommodates the HSA->ROC changes
* Adds instructions on installation and setup
* Updates examples and makes sure they work!
* Adds new device function example
* Adds section on ROC ufuncs, including async streaming gufuncs

Minor code changes:

* numba.roc.agents now works and lists the available agents

As title.