CUDA Greedy Resolver will break with more than 2^20 tracks

[beomki-yeo]

Currently scan_block_offsets kernel will have invalid arguments if there are tracks more than 10^20. There is also assertion to prevent this:

traccc/device/cuda/src/ambiguity_resolution/greedy_ambiguity_resolution_algorithm.cu

Line 356 in 8ba0a49

assert(nBlocks_scan <= 1024 &&

I was too lazy to handle this edge case, but will eventually fix it when I have time in the future. This issue is just not to forget about it.