Update server.py

Optimize CPU inference: fp32 instead of fp16, cgroup-aware threads

Rename spellchecker and normalize apostrophes

Create polished LFM2.5 spellchecker demo