Add POD version of TPCFastTransform

[shahor02]

The TPCFastTransformPOD is a pointerless version of the TPCFastTransform. It can be created from the original TPCFastTransform as e.g.

auto lold = o2::gpu::TPCFastTransform::loadFromFile("o2-gpu-TPCFastTransform.root","ccdb_object"); // load original transform 
std::vector<char> v; // one has to provide a vector (could be a std or pmr), which later can be messaged via DPL 
auto* pod = o2::gpu::TPCFastTransformPOD::create(v, *lold); // pointer pod is just v.data() cast to TPCFastTransformPOD*

// run test:
pod->test(*lold);
[INFO]  (ns per call)              original        this     Nmissmatch
[INFO] getCorrection               1.330e+02    1.400e+02   0
[INFO] getCorrectionInvCorrectedX  8.856e+01    8.434e+01   0
[INFO] getCorrectionInvUV          6.266e+01    6.142e+01   0

It can also be created directly from the TPCFastSpaceChargeCorrection as TPCFastSpaceChargeCorrection& oldCorr = lold->getCorrection(); auto* pod = o2::gpu::TPCFastTransformPOD::create(v, oldCorr); but in this case, one should afterwards set the vdrift and t0 using provided getters.

TPCFastTransformPOD replicates all the methods of the TPCFastTransform (and of the TPCFastSpaceChargeCorrection), including those which allow to query rescaled corrections (by providing reference maps and scaling coefficients). Since the idea of this class is to create a final correction map as a weighted sum of different contributions and to distribute it to consumer processes via shared memory, also the query methods w/o rescaling are added, they have the suffix _new added. Eventually, the scalable legacy methods can be suppressed, and the suffix new can be dropped.