Commit the work that has been done so far:

• Switch the way we measure the time in callout(9) from ticks to bintime.
• Change some fields in the structures callout, callout_cpu and cc_mig_ent.
• callout_reset_on() now converts the argument to_ticks to some bintime

value before giving it to callout_cc_add(), which has been modified to
receive a struct bintime rather than a value in ticks.

• Add a new per-cpu list to handle pending callout requests rather than

processing them directly on the callwheel. This simplify the logic of
the softclock() handler.

• Refactor the callout_tick() and callout_tickstofirst() code to make them

aware of the bintime measurement.

• Workaround tcp_timer_to_xtimer() to survive to the aforementioned switch.
• Refactor the cpu_new_callout() function in order to update the timer in

one-shot fashion via loadtimer() function. Until now the timer was
reprogrammed every tick.

Discussed with: mav