The random_sort function was only introducing randomness at the comparison
level rather than properly shuffling groups of records with the same position.
This meant that records with identical positions would maintain their original
relative order instead of being randomly distributed within their group.

Changes:
- Replace random_sort/random_compare with shuffle_records_by_position
- Group records by position before shuffling within each group
- Add comprehensive tests including edge cases and regression tests
- Fix clippy format string warnings in alignment.rs, fastx.rs, and subsampler.rs
- Add demonstration test that reproduces the original bug

The new implementation ensures that records with the same genomic position
are properly randomized while maintaining correct positional ordering between
different positions, fixing the reproducible bias in subsampling results.

Fixes #76