fix(e2ee): scope public-key cache by mid to prevent cross-contact key mixups

[nezumi0627]

Summary

The per-user E2EE public key cache was keyed only by keyId (e2eePublicKeys:${keyId}), not by mid.

LINE E2EE key ids are small per-account counters — each account's first registered key typically gets id 1 or 2. With two or more E2EE-capable contacts, the cache slots collide:

• Decrypting a message from contact B returned contact A's cached public key (no RPC was even attempted for B), producing a wrong ECDH shared secret and a GCM auth failure.
• Each successful negotiateE2EEPublicKey overwrote the shared slot, so two contacts with the same keyId continuously evicted each other's keys.
• The group path is affected too, since the creator key is fetched through the same function.

This also masked itself on the send side: encryptE2EEMessage calls negotiateE2EEPublicKey directly and never touches this cache.

Fix

Cache key becomes e2eePublicKeys:${mid}:${keyId} (read + write). Old keyId-only entries are simply never hit again and get re-negotiated once per contact — no migration needed.

Also fixes a latent always-throw path in the same block: when callers omitted keyId, negotiation succeeded but receiverKeyId == undefined was always false, throwing `E2EE key id undefined not found…` and wasting an RPC. An omitted keyId now means "accept whatever the server negotiated" (cached under its actual id).

Testing

New publickey_cache.test.ts:

1. Two contacts both with keyId 1: each lookup negotiates and returns that contact's own key material; repeated lookups still hit the per-mid cache (no extra RPC).
2. Omitted keyId: previously threw InternalError; now returns the negotiated key.

Full suite: deno test --allow-all — 216 passed, 0 failed.