Super sorter implemented

[lpgaff]

I have updated the time-ordering so that it doesn't fill a ROOT tree in the first instance, but a vector of the MiniballDataPackets class. These data are then time-ordered using the std::sort algorithm, which is super fast because it is done completely in memory. I tested the memory pressure and it is smaller than the histogrammer. We basically open the 2 GB MIDAS file in memory, then have all of that again in the vector, so we have something like 4 GB maximum memory usage. It did end up using something like 9 GB in the end, but there is no leak at least.

To write the output tree then, we loop over the sorted vector and fill the tree. Seems to work, but code needs some cleaning up. People are free to take a look at it for now though, might speed things up online.

I did a test with a run file from the IS708 run last year, results below for a subrun with 6525 x 64kB blocks of data.

ROOT standard sorting:

mb_sort -s settings_20241110.dat -c calibrations_20250304.dat -r  -i R9_14 -f 
94.80s user 3.54s system 99% cpu 1:38.86 total

New super sorter:

mb_sort -s settings_20241110.dat -c calibrations_20250304.dat -r  -i R9_14 -f
36.81s user 2.53s system 97% cpu 40.183 total

That is a factor of 2.5 times faster for small files, no -ebis flag included.

---

A larger 2 GB file with 32000 x 64 kB blocks of data gives:

ROOT standard sorting:

mb_sort -s settings_20241110.dat -c calibrations_20250304.dat -r  -i R9_13 -f 
426.61s user 11.70s system 99% cpu 7:19.73 total

New super sorter:

mb_sort -s settings_20241110.dat -c calibrations_20250304.dat -r  -i R9_13 -f
147.08s user 7.00s system 98% cpu 2:35.97 total

which is a factor of 2.9 times faster. So consistently better :-)