fix: bump IndexedDB version to invalidate old 256-dim cache

fix: defer index download to first search (lazy load)

Switch to potion-base-4M static embedding model

return 30 results instead of 20

compress data as much as possible

update

first commit