feat(memory): add bounded Memory entry pagination

docs/en/docs/operate/index.md

@@ -11,6 +11,7 @@ Start with the operational task you need:
 | --- | --- |
 | Keep a personal Server running, deploy a container, or enable authentication | [Deploy the Server](deploy-server.md) |
 | Upgrade or switch background supervision mode | [Migrate Artifact processing state](artifact-processing-migration.md) |
+| Enable bounded Memory-entry queries after upgrading an existing database | [Migrate the Memory query index](memory-query-migration.md) |
 | Inspect logs, metrics, and traces | [Observability](observability.md) |
 | Send traces to an analysis service | [Phoenix](trace-with-phoenix.md), [Langfuse](trace-with-langfuse.md) |
 | Diagnose failures and restore service or data | [Troubleshoot](troubleshoot.md) |

docs/en/docs/operate/memory-query-migration.md

@@ -0,0 +1,59 @@
+---
+title: Migrate the Memory query index
+---
+
+The bounded Memory-entry query uses a rebuildable directory projection. A new
+database is marked ready automatically. After upgrading a database that already
+contains Memory revisions, the Server keeps legacy list, exact-detail, search,
+and write operations available, but the new query returns
+`memory_query_index_unavailable` until this migration is verified.
+
+## Plan and apply
+
+Use the same deployment environment file as the Server:
+
+```shell
+powercontext server memory-query-migrate --action plan --env-file .env
+```
+
+The plan is read-only and reports missing projection tables, authoritative
+Memory revision count, existing directory rows, and the durable phase.
+
+Back up the database. Stop every old Server and Memory writer, disable their
+automatic restart, and keep them stopped for the maintenance window. The
+command cannot establish those external conditions. Then run:
+
+```shell
+powercontext server memory-query-migrate --action apply --env-file .env \
+  --migration-id rfc1656 --batch-size 100 --maintenance-confirmed
+powercontext server memory-query-migrate --action verify --env-file .env
+```
+
+`apply` creates any missing feature tables one at a time and commits at most the
+selected number of Memory revisions per backfill transaction. It verifies the
+result before reporting success. If it stops, repeat it with the same migration
+ID. Do not clear the marker or directory rows; a different ID is rejected while
+an earlier migration is incomplete.
+
+The batch size counts revisions, not entries inside a revision. Rebuilding one
+large initial manifest therefore writes one derived directory row per entry in
+that transaction. Those rows use set reads and executemany writes in chunks of
+500; choose the revision batch size and maintenance window with the largest
+stored manifest in mind.
+
+Only restart writers and Server replicas after the command reports
+`ready: true`. The separately runnable `verify` action rechecks every
+authoritative revision and tag-generation row and is safe to repeat.
+
+## What the migration changes
+
+The migration reads immutable Memory manifests and rebuilds revision-valid
+compact directory rows plus one tag generation per Memory Artifact. It never
+rewrites Memory Artifacts, entry versions, tags, or entry bodies. The directory
+remains derived state and is not used as an authority until full verification
+succeeds.
+
+Memory writes from the new version maintain their own directory deltas, but a
+rolling deployment with old and new writers during this offline migration is
+unsupported. SQLite migration requires a persistent database; in-memory SQLite
+has no upgrade state to preserve.

docs/en/docs/operate/meta.json

@@ -5,6 +5,7 @@
     "deploy-server",
     "connect-remote-server",
     "artifact-processing-migration",
+    "memory-query-migration",
     "observability",
     "trace-with-phoenix",
     "trace-with-langfuse",

docs/en/docs/operate/troubleshoot.md

@@ -269,7 +269,7 @@ so the previous database remains available for recovery:
 
    ```bash
    obloader <connection-options> -D <new-database> --csv \
-      --table 'pc_scopes,pc_source_journal_heads,pc_sources,pc_artifacts,pc_source_cursors,pc_artifact_processing_leases,pc_artifact_processing_binding_states,pc_artifact_processing_pending,pc_artifact_processing_auto_wave_targets,pc_artifact_processing_sequences,pc_artifact_processing_intents,pc_topic_memory_processing_targets,pc_artifact_processing_schema,pc_artifact_processing_migration_receipts,pc_topic_memory_work_budgets,pc_topic_memory_retrieval_shape,pc_connector_checkpoints,pc_source_definition_manifests,pc_external_skill_registrations,pc_skill_packages,pc_agent_skill_targets,pc_skill_publications,pc_model_usage_daily,pc_recall_token_daily,pc_receipt_migration_review' \
+      --table 'pc_scopes,pc_source_journal_heads,pc_sources,pc_artifacts,pc_memory_query_index_schema,pc_source_cursors,pc_artifact_processing_leases,pc_artifact_processing_binding_states,pc_artifact_processing_pending,pc_artifact_processing_auto_wave_targets,pc_artifact_processing_sequences,pc_artifact_processing_intents,pc_topic_memory_processing_targets,pc_artifact_processing_schema,pc_artifact_processing_migration_receipts,pc_topic_memory_work_budgets,pc_topic_memory_retrieval_shape,pc_connector_checkpoints,pc_source_definition_manifests,pc_external_skill_registrations,pc_skill_packages,pc_agent_skill_targets,pc_skill_publications,pc_model_usage_daily,pc_recall_token_daily,pc_receipt_migration_review' \
      -f <export-directory>
    ```
 
@@ -285,7 +285,7 @@ so the previous database remains available for recovery:
 
    ```bash
    obloader <connection-options> -D <new-database> --csv \
-      --table 'pc_artifact_candidate_heads,pc_topic_memory_active_topics,pc_topic_memory_active_chunks,pc_memory_entry_heads,pc_artifact_tags,pc_recurrence_match,pc_recurrence_observation' \
+      --table 'pc_artifact_candidate_heads,pc_topic_memory_active_topics,pc_topic_memory_active_chunks,pc_memory_entry_heads,pc_memory_entry_directory,pc_memory_tag_generations,pc_artifact_tags,pc_recurrence_match,pc_recurrence_observation' \
      -f <export-directory>
    ```