feat(engine): typed WaId + persistent lid->phone resolution table + message from-filter

[tobiasstrebitzer]

Closes #349.

Typed WaId value object + a persistent, cross-session lid -> phone resolution table + a from filter on message history. On top of main @ v0.4.5.

Composes with the v0.4.5 Baileys work (not duplicates)

• fix(baileys): resolve @lid senders to a phone number (#362) #372 added recordKeyLidMappings(key) (inbound senderPn/participantPn). This PR lets it feed the durable table: the method calls addLidMappings, which now writes through to the persistent lid_mappings table - so fix(baileys): resolve @lid senders to a phone number (#362) #372's inbound capture now persists cross-session for free.
• fix(baileys): show saved/contact names in the Chats list (#369) #370's resolveContactName is merged with this PR's contact/chat id canonicalization in toNeutralChat (neutral @c.us id and resolved name).

Motivation

Lid resolution was non-deterministic and ephemeral: a lid sender could be unresolved (@lid) early and resolved (@c.us) later once the lid -> pn map populated, and that map lived only in one Baileys session's memory (lost on restart, not shared across sessions). So a phone-based from filter matched a DM but silently missed the same person's already-resolved group message. A durable, shared table closes that; a typed WaId makes "a lid whose phone we don't know" first-class in the type.

New

• src/engine/identity/wa-id.value.ts - the WaId value object over the wa-id.ts primitives. In-memory only; toNeutral()/toString()/toJSON() yield today's neutral string; three-valued refersToSamePerson() (matched / didn't match / couldn't tell).
• src/engine/identity/lid-mapping.entity.ts + src/database/migrations/1781200000000-AddLidMappings.ts - LidMapping on the data connection (portable Postgres + SQLite migration).
• src/engine/identity/lid-mapping-store.service.ts - boot-loads the table into an in-memory map, synchronous forward (getCached) + reverse (lidsForPhone) lookups, write-through remember().

Wired (no contract changes)

• The store threads EngineFactory -> BaileysPlugin -> adapter config -> BaileysSessionStore, mirroring the existing messageStore seam.
• addLidMappings / upsertContacts write through to the table; resolvePhone falls back to the table's in-memory cache (kept synchronous).
• GET /api/sessions/:sessionId/messages?from= resolves through the table - a phone matches @c.us/@s.whatsapp.net stored ids and any lid resolving to it.
• Baileys contact/chat listing ids now emit @c.us (the feat(engine): engine-neutral WhatsApp identities (Baileys inbound conformance) #342-domain conformance that was deferred), merged with fix(baileys): show saved/contact names in the Chats list (#369) #370's name resolution; read-back lookups fold a neutral id back to the engine dialect so the round-trip still hits.

Acceptance criteria (from your review)

• Response shapes stay byte-identical. WaId is internal; a spec asserts WaId.toNeutral() equals toNeutralJid() for every representative id.
• Resolution table persists on the data connection. Entity on data, portable migration, boot-verified.
• From-filter test - a lid-resolved match becomes a hit. With the table mapping lid 111 -> 628999, filtering from=628999 returns the lid-authored row; a control test shows the prior silent miss.

Settled open questions

1. Wire/storage format = neutral string, WaId in-memory. ✅
2. 6.7.23 has no signalRepository.lidMapping lid->phone lookup, so unknown lids degrade to "couldn't tell" (@lid); resolution is fed by passive sources (inbound senderPn/participantPn via fix(baileys): resolve @lid senders to a phone number (#362) #372, phoneNumberShare, contacts, history sync). ✅
3. RESOLVE_LID_TO_PHONE resolves internally into the table and gates only what's exposed (privacy flag, not a correctness toggle). ✅

Deferred (issue #349 scope steps 3-6)

Module-by-module migration of consumers onto WaId (behind the existing string fields); active, lazy per-miss lid->phone lookup. The lid key-flip limitation you're tracking under #349 is addressed at the foundation (durable, cross-session identity); re-keying the specific in-memory consumers is part of the deferred migration.

A follow-up PR (feat/baileys-sync-hardening) stacks Baileys sync hardening on top of this.

[rmyndharis]

Thank you for this — it's a genuinely thorough piece of work, and I appreciate how carefully you staged the persistence and kept resolvePhone synchronous. I reviewed it closely against the source (migration portability on both SQLite and Postgres, the fire-and-forget write-through with its dedup, the boot-loaded cache, and the from-filter), and it all holds up.

I hope you don't mind — I pushed three small maintainer touch-ups directly to the branch so we could land it without a round-trip:

• The EngineFactory specs were still constructing with 3 args after you added the LidMappingStore 4th parameter, so that new wiring was being exercised as undefined (it slips past CI because specs aren't type-checked). They now pass a LidMappingStore fake.
• DROP INDEX/TABLE IF EXISTS in the migration down(), so a rollback is safe even when the table was created by the synchronize path (auto-named index).
• A small migration spec mirroring the sibling AddBaileysStoredMessages test.

Everything else is yours, and the design decisions are unchanged. One note for release framing: the contact/chat listing-id flip to @c.us is the consumer-visible piece we deferred in v0.4.5 — you've already documented it correctly in the CHANGELOG; we'll just settle patch-vs-minor at tag time.

Merging now. Thanks again for the careful work — this is a strong foundation for the identity arc.

[tobiasstrebitzer]

Thanks @rmyndharis appreciate your additional fixes and kind words