Improve performance

- [ ] look into a faster undordered map - for big files most of the time is spent on hashing and looking up in maps

- [ ] use a thread pool to do all of the exporting in paralel

- [ ] maybe use std terminate to end program execution earlier so we don't have to wait for all those strings and maps to be freed?

Start by profiling analysis of big files to confirm how much time hashing and map lookups consume. Then investigate the exporting path for safe thread-pool parallelism and assess whether earlier termination is appropriate; done means agreeing on a focused approach and measuring a demonstrated performance improvement.

Written by the indexing model from the issue text.