fix: tool start hanging + SSE 429 errors

1. Remove Docker build from create_instance for manifest types — the build
   was blocking the HTTP request for several minutes, causing frontend
   timeouts and retries. Image is now built lazily on start (via the
   existing _prepare_manifest_instance path in start_instance).

2. Increase MAX_CONNECTIONS_PER_USER from 5 to 20 for SSE endpoint —
   aggressive reconnect loops from the frontend were exhausting the limit
   and causing 429 errors unrelated to tool starting.

Quality gates: ruff clean, tsc --noEmit clean, pytest workspaces (9 passed)

apps/api/src/api/events.py

@@ -16,7 +16,7 @@ router = APIRouter(prefix="/events", tags=["events"])
 
 # In-memory connection counter per user (single-process assumption)
 _connection_counts: dict[uuid.UUID, int] = {}
-MAX_CONNECTIONS_PER_USER = 5
+MAX_CONNECTIONS_PER_USER = 20
 
 
 @router.get("/stream")

apps/api/src/api/tool_instances.py

@@ -1039,7 +1039,7 @@ services:
             write_compose_file(instance_dir, compose_content)
 
         elif tool_type.definition_type == "manifest":
-            # Manifest-based: build image and generate compose
+            # Manifest-based: generate compose only; image built lazily on start
             from src.models.tool_definition_manifest import ToolDefinitionManifest
 
             manifest_def = await session.get(
@@ -1061,44 +1061,8 @@ services:
                         deep_merge(dict(base_def.manifest), manifest)
                     )
 
-            # Determine home directory for path expansion
-            _home_dir = get_manifest_home_dir(manifest)
-
             image_tag = compute_image_tag(tool_type.name, manifest)
 
-            # Build image during creation so start is fast
-            dockerfile = compile_dockerfile(manifest)
-            entrypoint = compile_entrypoint(manifest)
-            build_ctx = {
-                "Dockerfile": dockerfile,
-                ".headquarter/entrypoint.sh": entrypoint,
-            }
-
-            returncode, stdout, stderr = await asyncio.to_thread(
-                build_image,
-                instance_dir=instance_dir,
-                dockerfile=dockerfile,
-                tag=image_tag,
-                build_context=build_ctx,
-            )
-
-            if returncode != 0:
-                logger.error(
-                    "Failed to build image for manifest instance %s: %s",
-                    instance_name,
-                    stderr,
-                )
-                raise HTTPException(
-                    status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-                    detail=f"Failed to build Docker image: {stderr[:500]}",
-                )
-
-            logger.info(
-                "Built manifest image %s for instance %s",
-                image_tag,
-                instance_name,
-            )
-
             variables = {
                 "IMAGE_TAG": image_tag,
                 "INSTANCE_NAME": instance_name.lower(),

apps/api/src/services/terminal_manager.py

@@ -91,14 +91,18 @@ class TerminalManager:
         """
         try:
             async with SessionLocal() as db_session:
-                stmt = pg_insert(TerminalSessionModel).values(
-                    id=uuid.UUID(session_id),
-                    instance_id=instance_id,
-                    name=name,
-                    status="active",
-                    created_at=datetime.now(timezone.utc),
-                    last_activity_at=datetime.now(timezone.utc),
-                ).on_conflict_do_nothing(index_elements=["id"])
+                stmt = (
+                    pg_insert(TerminalSessionModel)
+                    .values(
+                        id=uuid.UUID(session_id),
+                        instance_id=instance_id,
+                        name=name,
+                        status="active",
+                        created_at=datetime.now(timezone.utc),
+                        last_activity_at=datetime.now(timezone.utc),
+                    )
+                    .on_conflict_do_nothing(index_elements=["id"])
+                )
                 await db_session.execute(stmt)
                 await db_session.commit()
                 logger.debug(