refactor(layers): flip the output seam — view bridge moves into features/feed (#493 PR-3) - #503

…res/feed (#493 PR-3)

rendering/view/ ceases to exist:
- ledgerFeedItems.ts (+ test) and useLedgerFeedItems.ts move to
  features/feed/ledger/. Every one of the bridge's imports is now legal:
  FeedRenderItem / VisibleDecision / groupSemanticActivity are same-layer
  siblings; RenderLedger/RenderRow come from rendering/model (feed ->
  rendering, the sanctioned direction). The react import leaves rendering/
  entirely — the decide layer is now framework-free.
- groupSemanticActivity STAYS feed-side (plan D3): issue PR-2 wanted it in
  rendering/view, but with the whole view stage relocating here that would
  have been round-trip churn; presentation grouping is render-layer.

Replay harness un-inverted by injection:
- recordedSession.ts no longer imports the bridge. New seam:
  ReplayOptions.projectItems (ReplayItemsProjection) — the harness runs
  slices -> adapter -> ledger and stops at RenderLedger, the decide
  layer's own output; projecting rows into feed items is the render
  layer's job, injected by callers. ReplayTick.feedItems is unknown[]
  (empty when no projector given).
- recordedSession.test / invariants.test / recordingCorpus.test inject the
  REAL production bridge from features/feed/ledger, so full-stack coverage
  (incl. the #239 dropped-candidate accounting) is unchanged — test files
  are outside the layer contract by design.

The #493 acceptance now holds and is grep-verified:
  grep -rn "@renderer/features|@renderer/workspace|from 'react'" \
    src/renderer/src/rendering/ --include='*.ts' | grep -v '\.test\.'  # -> empty
  grep -rn "@renderer/(workspace|features|app-state)|from 'react'" \
    src/renderer/src/session-runtime/                                  # -> empty

Three folders, one-way: session-runtime (ingest) <- rendering (decide) <-
features/feed (render); the ledger consumes exactly RuntimeRenderInput and
emits exactly RenderLedger/RenderRow[].

Gates: tsc node+web clean, vitest 467/467, remote-client build OK. Stack
is linear on origin/main (fast-forward merge).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…rojection, refresh moved-path comments

Two Claude + two Codex reviewers over the full stack; zero blockers. The
two actionable findings:

- projectItems is now a REQUIRED ReplayOptions field (codex finding): an
  omitted projection was indistinguishable from "projected with zero
  drops", silently vacating the #239 dropped-candidate invariant for that
  run. Required-field = the forgotten-injection mistake is a compile
  error, not a green-but-hollow test.
- Stale comment path pointers to pre-#493 module locations refreshed
  (claude finding): reconstructSlices mirror-source block, Feed.tsx ghost
  notes, streamingWriteInput, observations/{ghosts,semantic},
  providerConfig, main/{ghostJournal,index}. Navigation pointers are
  exactly what the CLAUDE.md comment policy exists to keep accurate.

Codex's remaining finding — the phone's hard-coded lastJsonlEntryAt: 0 —
pre-exists this stack (the old cast hid the same value) and needs real
committed-tail plumbing on the wire protocol; deferred to a follow-up
rather than smuggled into a behavior-preserving refactor.

Gates: tsc node+web clean, vitest 467/467.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>