fix(memory-sync): reliable per-source sync indicators and counters

[graycyrus]

Summary

Fixes the unreliable per-source "Syncing…" indicators and the live "N syncing / N synced" counters on the Intelligence memory sources panel (MemorySourcesRegistry). Previously only one source could show as syncing at a time, intermediate stages never lit up the per-source bar, and progress got stuck at a placeholder.

Closes #3295.

Root causes addressed

1. Only one source could show "syncing" — MemorySourcesRegistry.tsx tracked a single syncingId: string | null. Now a syncingIds: Set<string> so concurrent syncs each render their own progress (immutable Set updates).
2. Intermediate stages carried the wrong id — MemorySyncStageBridge (src/openhuman/memory/sync.rs) emitted stored/queued with no connection_id and ingesting with connection_id = document_id, so the frontend (matching by connection_id === source.id) never lit up intermediate stages. Added a new optional source_id: Option<String> field to MemorySyncStageChanged (distinct from connection_id, which still carries the ingest-pipeline document_id). The bridge extracts the originating memory-source id from the mem_src:<source_id>:<item_id> chunk encoding (first-colon split, since item ids may be URLs). Frontend matches on source_id ?? connection_id (backward compatible).
3. Brittle progress parsing — parseSyncProgress now tolerates non-N/M detail strings and falls back to per-stage default percentages so the bar always advances.
4. Event/poll races — the 5s poll reconciles missed events to a done state; documented the minor remount gap.

Out of scope (intentional, tracked separately)

• Root cause Feat/landing revamp #3 (two accounts of the same Composio toolkit collapsing into one counter row) — requires reworking Composio chunk tagging with data-migration implications; deferred to a follow-up.
• Root cause sync branches #5 full recovery — no new memory_sources_active_syncs RPC; rely on the existing 5s poll to rehydrate after a remount.

Event payload change

MemorySyncStageChanged gains source_id: Option<String> (the memory-source row id; None for channel-level syncs). connection_id semantics are unchanged. Serialized into the memory-sync-stage socket payload.

Tests

• Rust (src/openhuman/memory/sync.rs): extract_mem_src_id unit tests (incl. URL item ids with colons) + MemorySyncStageBridge tests asserting source_id is populated for mem_src: syncs and None otherwise, with connection_id left unchanged. 10 lib tests pass.
• Vitest (app/src/components/intelligence/__tests__/MemorySourcesRegistry.test.tsx): concurrent syncs render both rows, completed/failed clears only the right row, source_id preferred over connection_id, and tolerant parseSyncProgress cases. 11 tests pass.

Checks

pnpm typecheck, pnpm lint (0 errors), pnpm format:check, pnpm build, cargo check, the new Vitest suite, and the memory::sync Rust lib tests all pass. The full pnpm test:rust e2e suite and pnpm dev:app were not run locally due to a disk-space limit in the dev environment (ld: errno=28); the library tests covering this change pass.

Notes for reviewers

• connection_id must keep carrying document_id for ingest-pipeline identity — this PR adds source_id rather than repurposing connection_id. Please sanity-check no downstream consumer depends on connection_id being the source id.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 808b08ac-511e-4373-9193-002c13e12b78

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[CodeGhost21]

Review withdrawn — PR is in draft. Will re-post once it's ready for review.