30% faster dataset validation in Soup

People training models on laptops just got 30% faster dataset validation on Soup.

The validator checks your datasets before training starts. I optimized it to run in a single pass instead of multiple passes.

Changed row serialization to hashable row signatures. Removed intermediate allocations.

20,000 rows: 0.1397s → 0.0971s.

For users, nothing changed. Same commands. Same configuration. Just less time waiting.

The PR is merged and shipped in Soup v0.75.0. PR: MakazhanAlpamys/Soup #771. Release: v0.75.0.