sampler_poisson: Fix crash due to vector length being hardcoded to 1M

[nh2]

Fixes #550.

[nh2]

Note that while this fixes the segfault, on one of my datasets that previously crashed I now got out-of-memory. I don't know if that is explained by the extra vector entries that are now allowed.

For another, 20x larger dataset las files of total of > 2000 GB, it did not crash nor OOM, so I wonder in which case this code path is even triggered.

I don't fully understand the meaning/purpose of dbgAccepted yet; unfortunately the code has no comments that explain what is done and why.

[nh2]

I don't fully understand the meaning/purpose of dbgAccepted yet; unfortunately the code has no comments that explain what is done and why.

I now understood it after reading the PotreeConverter paper at https://www.cg.tuwien.ac.at/research/publications/2020/SCHUETZ-2020-MPC/SCHUETZ-2020-MPC-paper.pdf

while this fixes the segfault, on one of my datasets that previously crashed I now got out-of-memory

It was because of my change. I didn't notice the thread_local qualifier on the dbgAccepted vector.

I've pushed a fix to this overuse of memory my PR introduced so far.