GPU Metal: run the tracklet constructor with 128 threads per group

The tracklet constructor needs the most registers a Metal thread can
have, which leaves room for 384 threads on a core of an M1 Max, so with
256 threads per group only one group fits. With 128 threads per group
two to three groups do, and the kernel takes 20% less time: 516 ms ->
415 ms on a quarter of a 30 kHz Pb-Pb time frame (--debug 1, interleaved
runs), with the same number of tracks. 96 and 192 threads are within
2% of 128; 64, 160, 256 and 384 are slower. The kernel loops over its
start hits with a grid stride and its shared memory does not depend on
the group size.

Author: ktf

GPU/GPUTracking/Definitions/Parameters/GPUParameters.csv

@@ -6,7 +6,7 @@ THREAD_COUNT_DEFAULT,256,,256,256,,,,,,512,512,,512,256,512,512,512,,256
 ,,,,,,,,,,,,,,,,,,,
 LB:,,,,,,,,,,,,,,,,,,,
 GPUTPCCreateTrackingData,256,,"[256, 7]","[192, 2]",,,,,,"[224, 7]",256,"[128, 14]",416,,"[64, 21]",,384,"[320, 2]",64
-GPUTPCTrackletConstructor,256,,"[768, 8]","[512, 10]","[256, 2]","[256, 1]","[256, 2]","[1024, 2]","[512, 4]",1024,"[256, 2]",1024,"[1024, 1]",,"[768, 2]",,768,512,"[256, 32]"
+GPUTPCTrackletConstructor,256,,"[768, 8]","[512, 10]","[256, 2]","[256, 1]","[256, 2]","[1024, 2]","[512, 4]",1024,"[256, 2]",1024,"[1024, 1]",,"[768, 2]",,768,512,"[128, 32]"
 GPUTPCTrackletSelector,256,,"[384, 5]","[192, 10]","[256, 3]","[256, 1]","[256, 3]","[512, 4]","[256, 3]","[288, 3]","[192, 3]","[544, 1]","[32, 2]",,"[384, 3]",,992,"[256, 6]",
 GPUTPCNeighboursFinder,256,,"[192, 8]","[960, 8]",256,256,256,512,256,864,"[640, 1]","[512, 2]","[736, 1]",,"[480, 3]",,992,"[704, 1]",
 GPUTPCNeighboursCleaner,256,,"[128, 5]","[384, 9]",256,256,256,256,256,544,512,"[192, 9]","[512, 1]",,"[384, 5]",,672,"[640, 1]",