Drop unused agent_messages.read_at column via migration v49 (M1)

The `agent_messages.read_at` column has had no production writer since
v12 moved read tracking into the per-viewer `agent_message_reads`
receipts table. It survived anyway, carrying a NULL through every SELECT
list and begging the next maintainer to accidentally read stale data
from it.

- v49 migration: `ALTER TABLE agent_messages DROP COLUMN read_at`
  (idempotent: skips if absent).
- messages.ts: SELECTs no longer fetch the column; per-viewer `readAt`
  is computed from `agent_message_reads` as before, with `null` for
  call-sites that have no viewer context (getAgentMessage /
  searchMessages / listAgentInboxUnread).
- runtime-bridge.ts / runtime-message-injector.ts: comments referencing
  the dropped column updated to name the read-receipts table instead.

src/main/agent-runtime/runtime-bridge.ts

@@ -562,16 +562,16 @@ export class RuntimeBridge {
     // as plain user turns in its context.
     //
     // Replay safety lives entirely on the `agent_message_projections`
-    // ledger. We deliberately do NOT touch `agent_messages.read_at`
-    // here even though that column exists: "we put the message into
-    // the agent's transcript" and "the agent has acknowledged the
-    // message" are two different facts, and welding them onto one
-    // column made every projected message immediately invisible to
-    // the agent's `?unreadOnly=true` API call. The agent then read
-    // its IC prompt ("respond to ALL unread messages"), called the
-    // API, got an empty list, and posted a status comment instead of
-    // replying. The explicit POST /messages/{id}/read is now the
-    // sole signal that flips `read_at`.
+    // ledger. "We put the message into the agent's transcript" and
+    // "the agent has acknowledged the message" are two different
+    // facts, tracked in two different tables: projections are append-
+    // only on dispatch, while reads are inserted into
+    // `agent_message_reads` only when the agent POSTs
+    // /messages/{id}/read. Conflating the two (as an earlier build
+    // that wrote to `agent_messages.read_at` at projection time did)
+    // made every projected message immediately invisible to
+    // `?unreadOnly=true`, so agents posted status comments instead of
+    // replies. The explicit read endpoint is the sole signal.
     const store = this.getOrCreateStore(payload.taskSessionId);
     const inboxRecords = this.deps.db.listAgentInboxUnread(payload.companyId, payload.agentId);
     if (inboxRecords.length > 0) {

src/main/agent-runtime/runtime-message-injector.ts

@@ -134,10 +134,10 @@ export class RuntimeMessageInjector {
    * the recipient is woken so the next `prepareRun` picks the message
    * up through the regular inbox projection.
    *
-   * `read_at` is intentionally NOT touched here: "the worker queued
-   * this message onto the agent" is not "the agent replied". The
-   * agent flips `read_at` itself by POSTing /messages/{id}/read once
-   * it has actually processed the row.
+   * The read receipt is intentionally NOT written here: "the worker
+   * queued this message onto the agent" is not "the agent replied".
+   * The agent inserts into `agent_message_reads` itself by POSTing
+   * /messages/{id}/read once it has actually processed the row.
    */
   handleSteerAck(correlationId: string, accepted: boolean): void {
     this.resolvePending(correlationId, accepted, "worker_ack");