ui: merge duplicate per-cpu/gpu tracks on multi-machine open

A merged trace can contain several track rows for one logical entity:
two traces on the same machine each intern their own cpu_frequency,
gpufreq, cpu-per-uid totals and gpu work-period tracks. Keying a track
URI off a single row's dimension made registerTrack() throw
RegistryError on the duplicate; keying it off the row id avoided the
crash but split one CPU/GPU into several tracks.

Follow the Sched plugin: enumerate the logical entity (grouping by
machine plus its machine-unique key) and have the single resulting track
query across all of that entity's track ids. This removes the collision
and unions the rows into one contiguous track, keeping distinct machines
separate.

  CpuFreq       group by ucpu; union freq/idle track ids
  Gpu frequency group by (machine, ugpu|gpu_id); union track ids
  GpuWorkPeriod group by (machine, ugpu|gpu_id, uid); union track ids
  CpuPerUid     group by (machine, type, cluster); union track ids

Sched, CpuidleTimeInState, CoarseCpu, LinuxPerf and the Gpu
counter/slice tracks already enumerate distinct keys or suffix the track
id, so they are unaffected.

Change-Id: Ic7e13738f6fc1b2229f9bd790e6be7fc8dc94d20