The app was computing MI (mutual information) metrics but not surfacing
them to users. This change makes MI similarity data visible and
actionable throughout the UI, enabling users to answer the core
question: "How does genetic similarity compare to taxonomy?"

Backend:
- Add mi_norm and align_len to GraphEdge schema (was only distance)
- Add shared_rank to NeighborOut (computed from lineage arrays)
- MI network deduplication now preserves mi_norm and align_len
- Neighbors endpoint batch-fetches lineage ranks for taxonomy coherence

Frontend:
- NeighborCard shows actual NMI% instead of misleading relative distance
- Neighbor cards show shared taxonomic rank badge (genus/family/order)
- Taxonomic coherence summary bar above neighbors section
- Graph page MI stats panel (avg/median/range NMI, distribution buckets)
- Sigma graph edge hover tooltip showing NMI%, distance, alignment cols
- TypeScript types updated for new GraphEdge and NeighborOut fields

Tests: 122 API tests + 86 frontend tests passing (5 new tests added)

https://claude.ai/code/session_015FCj2G9CHpnsJqHcF3n8aw