Uorb optimization3

[jlaitine]

This changes the way how uORB callbacks are executed;

• The callback subscriptions are collected in a list withing each process
• The priority of the processes' callback thread is set to the maximum thread priority within the process
• All the triggred callbacks are executed in a loop when the callback thread gets to run

Signalling the callback

• The publisher marks the callback as "triggered" and signals the callback thread. There is no synchronization back from the callback thread potentially allowing the publisher to fire all the registered callbacks/polls at once withouth scheduling away at each signalling semaphore post.

[jlaitine]

Pushed one more patch to orb_poll, which shaves off 1pp total execution time, from uxrce_dds client.

The callback optimization shave off ~7pp

[jlaitine]

just re-based to see if also the sitl tests pass now

[pussuw]

LGTM!

[jlaitine]

There is something wrong with this; it just doesn't fly properly. Need to investigate more

[jlaitine]

Did several cleanups, now it is flying good.

The CPU savings are not so big as I hoped, but still around 5pp . But I think that the architecture around callbacks is also cleaner this way, so quite happy with this version.

I removed the orb_poll changes and the dynamic priority changing; they just didn't prove effective.

[jlaitine]

Added some minor cleanups, tested again on default build, kernel build and sitl