Initial commit for new event union

[scottatchley]

This does not include any source changes.

[dillow]

I mostly approve, a few nits:

• Please watch out for trailing whitespace issues. 'git diff --color=auto' will highlight those in red for you.
• There are a number of typos:
  signalled => signaled
  cci_event_connect_req_t => cci_event_connect_request_t (or vice versa)
  cci_event_device_failed => cci_event_device_failed_t
  cci_event_connect_device_failed_t => cci_event_device_failed_t
  etc.

Please fix these and rebase; I'd prefer to not have a long chain of fixups in the history.

[dillow]

Also, check out the duplicity between cci_event_connect_request_t and cci_conn_req_t; I need the latter for accept/reject, but I don't have a way to get there from the event; perhaps we keep cci_get_conn_req() around, but then there is no link from event to the request.

We didn't want to embed the conn_req_t stuff in the event as we thought people might sit on them for a while -- LDAP auth, etc. So should we then have a pointer in the event that becomes a cookie for accept/reject() to use to get to the driver specific state?

[dillow]

We could also preallocate 'backlog' conn_req events, and just those as the key; we could copy any data needed out of the RX ring -- I'm assuming it is a maximum of 1 KB. This keeps use from hogging the AM resources, and merges the event.

[scottatchley]

Dave, re-whitespace issues, I have have been slowly cleaning up the header file as we touch related code. I set c_space_errors = 1 in vimrc, but I turned it off for the conf call. ;-)

I will fix and rebase the typos. This pull request was rushed. I should have waited.

There is so much overlap with the remove-header pull request, that we may be better off waiting until that one is landed and letting me start over.

[scottatchley]

I will close this request until the topic/remove-header pull request is landed to avoid overlap.