v0.4.0

What's new in v0.4.0

LLM & Translation

• Ollama health check before glossary extraction and translation — clear error message instead of cryptic TCP failure
• 1-retry with 2 s delay on OllamaProvider::chat() for transient network errors
• Request timeout raised from 120 s to 180 s for cloud Ollama (RunPod, etc.)
• Settings Test button now shows explicit success/failure toast; auto-calls (modal open, URL blur) remain silent

Extraction Quality

• Shared engines/filter.rs module — noise segments (pure numbers, pure symbols, pure placeholders) are now skipped across all engines (MV/MZ, VX Ace, Wolf)
• ~102 noise segments eliminated on reference MV game (-, …, ！！！！, etc.)
• Glossary auto-extraction now includes actor nicknames (/nickname keys) in addition to names

Docs

• Add Cloud LLM with RunPod section to README (EN + FR) — full pod setup, Start Command with model pull, recommended batch sizes per GPU
• Wolf RPG marked as ⚠️ Partial (.dat only)