CheckOverlap uses a significant amount of time in 2025 in HLT2
When running a throughput test on 2025 data, we realized that checkOverlap uses close to 10% of the full HLT2 sequence. This is caused by creating many combinations (documented here: Moore#967), but it also needs understand if it can be improved from the C++ side.
The flame graph was obtained with a modified MooreOnline_hlt2_pp_2024_data.sh script:
MooreOnline_hlt2_pp_2025_data.sh
using the hlt2_pp_2025 sequence.
This needs: lhcb-datapkg/PRConfig!550