feat(sync): add opt-in automatic multi-machine sync on top of existing chunk sync

[evelioapc]

📋 Pre-flight Checks

• I have searched existing issues and this is not a duplicate
• I understand this issue needs status:approved before a PR can be opened

🔍 Problem Description

Engram already has the right primitive for multi-machine usage: engram sync exports/imports compressed chunks that are safe to move between machines.

That works, but it is still manual:

1. Save memory on machine A
2. Run engram sync
3. Commit/push or otherwise transfer the sync artifacts
4. Pull on machine B
5. Run engram sync --import

For users who switch between two or more computers during the day, this friction is high enough that memory becomes stale in practice.

Important constraints:

• Syncing the live SQLite file through OneDrive/Dropbox/etc. is unsafe
• The current chunk-based sync model is correct, but the UX is incomplete for real multi-PC workflows

💡 Proposed Solution

Add an opt-in automatic sync layer on top of the existing chunk-based sync model.

Design goals:

• Keep SQLite local as the source of truth
• Keep the existing chunk export/import format
• Add background/scheduled auto-sync
• Make the transport pluggable so Git can be supported first, with Cloud as a future option

Proposed UX:

Configure a sync profile once:
engram sync init git --repo ~/engram-sync

Then start Engram normally:
engram mcp

Behavior:

• on startup: pull/import remote changes if available
• after local writes: debounce, export chunks, commit, push
• periodically: pull/import in the background
• on shutdown: final sync attempt

Suggested commands:

• engram sync init git --repo <path-or-url>
• engram sync enable
• engram sync disable
• engram sync status
• engram sync run-once
• engram sync doctor

Implementation outline:

• add a sync config file under ENGRAM_DATA_DIR
• add sync orchestration on top of existing export/import logic
• add Git transport hooks for pull, add/commit, and push
• never block writes if sync fails
• never sync the live SQLite database file directly

📦 Affected Area

Sync (multi-instance)

🔄 Alternatives Considered

1. Sync the SQLite file with Dropbox/OneDrive
   Rejected because it is unsafe with concurrent machine usage and risks corruption/conflicts.

2. Require PostgreSQL/shared DB for multi-machine use
   Rejected for this issue because it is too heavy for solo users with multiple machines and overlaps with separate shared-memory/team proposals.

3. Keep manual sync only
   Rejected because it technically works, but the UX friction is high enough that many users will not keep memory in sync across devices.

📎 Additional Context

Related:

• Existing manual Git sync in README/docs
• Engram Cloud Sync - Mutations Marked as Synced but Not Reaching Cloud #66 cloud sync reliability bug
• feat(store): add PostgreSQL backend for team shared memory with git identity tracking #150 PostgreSQL/team shared memory proposal

Why this matters:

Engram is already very close to solving the same-memory-on-every-machine problem.

The missing piece is not a new storage engine. It is a safe, opt-in automation layer built on top of the sync model that already exists.