Drop the mutexes and special struct, just create ring buffer of
GstClockTimes. Also rely on callbacks for start/stop of acquisition.
The absolute date-timestamps need to be corrected for the first buffer
timestamp.
Unfortunately IMAQ only lets you examine one buffer at a time, and elements
like queue store up multiple buffers, so avoid-copy=TRUE can only be used in
pipelines without queues or other such elements.
Also, make property strings static.
Previously if the row stride wasn't a multiple of 4, display would not work
correctly. Rely on NI-IMAQ to fix the byte alignment for us. This hasn't
been thoroughly tested yet however.
Even though NI-IMAQ can specify 10-, 12-, or 14-bit images, no other standard
element supports it, and in 0.11/1.0 caps are simplified to be gray8/gray16.
We can always add back custom caps if there is a real need for it.
Caused seg fault unless delay was introduced (e.g. with debugging messages
on). Still need to investigate why we don't get a timestamp before the
first frame.
