Decide: collapse the browser stream mirror first, or move it into the package first

Jonas's direction (2026-07-17 jam): the SQLite/feed plumbing should ALSO live in packages/iterate eventually — "use the same plumbing for populating a SQLite database and having feed items and all that jazz".

The tension: moving the mirror (~6k LOC movable core per the #2063-era sweep) drags the stream-processor / stream-processor-runner family with it (the browser hosts processors via the same runner the DOs use) — while apps/os/docs/replace-browser-stream-database.md (written during #2048) designs the OPPOSITE: server-owned feed live view + cursor-paged history, then DELETE ~4,850 LOC of the mirror. Moving code that is slated for deletion is churn.

This task is the DECISION, not the work: pick collapse-first (Phases A–D in the design doc, then move what survives) or move-first (accept the runner family relocating), and spawn the real tasks from the choice. Inputs: the collapse design doc, docs/frontend-development.md's "one exception" section, and PR #2063's next-steps list.