fix(monitoring): paginate conversation IDs before loading message payloads

Count and paginate lightweight records, then load payloads only for the
selected page to avoid materializing large historical messages.

Add regression coverage for oversized history, ingress deduplication,
stable pagination, and PostgreSQL tenant isolation.

src/langbot/pkg/api/http/service/monitoring.py

@@ -2060,7 +2060,7 @@ class MonitoringService(ExecutionDetailsMixin):
                 query = query.where(source.c.timestamp <= end_time)
             return await self._conversation_projection_page(
                 workspace_uuid,
-                query.order_by(source.c.timestamp.desc(), source.c.id.desc()),
+                query.order_by(source.c.timestamp.desc(), source.c.id.desc(), source.c._source),
                 limit,
                 offset,
             )

src/langbot/pkg/api/http/service/monitoring_execution_details.py

@@ -101,6 +101,19 @@ def transcript_item(row):
 
 class ExecutionDetailsMixin:
     def _conversation_message_source(self, workspace, bot_ids):
+        """Count and page identities without carrying historical payloads."""
+        fields = ('id', 'workspace_uuid', 'timestamp', 'bot_id', 'session_id')
+        return sa.union_all(
+            *[
+                query.with_only_columns(
+                    *[query.selected_columns[field] for field in fields],
+                    sa.literal(origin).label('_source'),
+                )
+                for origin, query in zip(('message', 'event'), self._conversation_message_queries(workspace, bot_ids))
+            ]
+        ).subquery()
+
+    def _conversation_message_queries(self, workspace, bot_ids):
         """Combine delivered messages with ingress missing from the legacy ledger."""
         message = models.MonitoringMessage
         conversation = sa.func.coalesce(
@@ -176,17 +189,47 @@ class ExecutionDetailsMixin:
                 )
             ),
         )
-        return sa.union_all(stored, incoming).subquery()
+        return stored, incoming
 
     async def _conversation_projection_page(self, workspace, statement, limit, offset):
         count = (
-            await self._execution_query(workspace, sa.select(sa.func.count()).select_from(statement.subquery()))
+            await self._execution_query(
+                workspace, sa.select(sa.func.count()).select_from(statement.order_by(None).subquery())
+            )
         ).scalar_one()
-        rows = (await self._execution_query(workspace, statement.limit(limit).offset(offset))).mappings().all()
-        return [
-            {key: value.isoformat() if isinstance(value, datetime.datetime) else value for key, value in row.items()}
-            for row in rows
-        ], count
+        page = (await self._execution_query(workspace, statement.limit(limit).offset(offset))).mappings().all()
+        if not page:
+            return [], count
+
+        # Fetch payloads only for identities selected by the bounded page.
+        queries = self._conversation_message_queries(workspace, list({row['bot_id'] for row in page}))
+        rows = {}
+        for origin, query in zip(('message', 'event'), queries):
+            keys = [(row['bot_id'], row['id']) for row in page if row['_source'] == origin]
+            if not keys:
+                continue
+            selected = query.selected_columns
+            result = await self._execution_query(
+                workspace,
+                sa.select(*selected).where(
+                    selected.workspace_uuid == workspace,
+                    sa.tuple_(selected.bot_id, selected.id).in_(keys),
+                ),
+            )
+            for row in result.mappings():
+                rows[(origin, row['bot_id'], row['id'])] = row
+
+        messages = []
+        for identity in page:
+            row = rows.get((identity['_source'], identity['bot_id'], identity['id']))
+            if row is not None:
+                messages.append(
+                    {
+                        key: value.isoformat() if isinstance(value, datetime.datetime) else value
+                        for key, value in row.items()
+                    }
+                )
+        return messages, count
 
     def _execution_json_text(self, column, key):
         # SQLite JSON functions accept TEXT directly; PostgreSQL operators