feat(agent): support Responses steering and project-aware tools

Use the dedicated TJWATER_AGENT_URL and Responses stream protocol. Keep running
turns active while users steer or queue follow-ups and edit pending delivery.
Refresh the model catalog before generation and preserve approval mode selection.

Integrate scoped investigation messaging/presentation, geographic place selection,
current Agent composer styles and supporting workbench contracts.

Validation: 77 focused Vitest checks, 21 browser cases, lint, typecheck and
production build. Preserve direct Server/Agent routing and project authentication.

src/features/workbench/api/operational-task-client.ts

@@ -1,5 +1,6 @@
 import { z } from "zod";
-import { investigationReportSchema } from "./investigation-report";
+import { investigationReportSchema, investigationDiagramSchema } from "./investigation-report";
+import { investigationPresentationCursorSchema } from "./investigation-presentation";
 import type { AccessTokenProvider } from "@/shared/auth/keycloak-auth";
 import { env } from "@/shared/config/env";
 import { createWorkbenchHeaders, findOwnedWorkbenchProject, getServerError } from "./project-client";
@@ -45,6 +46,7 @@ const taskSummarySchema = z.object({
   workflow: z.string().min(1),
   lifecycle_status: z.enum(["open", "waiting_human", "completed", "failed", "cancelled"]),
   active_run: operationalTaskRunSchema.nullable(),
+  latest_run: operationalTaskRunSchema.nullable().optional(),
   status: taskStatusSchema,
   priority: taskPrioritySchema,
   revision: z.number().int().positive(),
@@ -115,6 +117,14 @@ const reportRevisionSchema = z.object({
   published_by: z.string().min(1),
   report: reportInputSchema
 }).strict();
+const progressSnapshotSchema = z.object({
+  revision: z.number().int().positive(),
+  run_id: z.string().uuid(),
+  execution_generation: z.number().int().positive(),
+  published_at: workbenchDateTimeSchema,
+  source_session_id: z.string().min(1),
+  progress: z.object({ summary: z.string().trim().min(1), diagram: investigationDiagramSchema.nullable() }).strict()
+}).strict();
 const reportPageSchema = z.object({
   items: z.array(reportRevisionSchema),
   next_offset: z.number().int().nonnegative().nullable(),
@@ -212,6 +222,7 @@ const externalWorkOrderSnapshotSchema = z.preprocess((value) => {
 const operationalTaskSchema = taskSummarySchema.extend({
   latest_plan: planSnapshotSchema.nullable(),
   latest_report: reportRevisionSchema.nullable(),
+  latest_progress: progressSnapshotSchema.nullish(),
   evidence: z.array(evidenceSchema),
   actions: z.array(actionSchema),
   access_requests: z.array(accessRequestSchema).default([]),
@@ -269,7 +280,43 @@ const scheduleSchema = z.object({
 }).strict();
 const schedulePageSchema = z.object({ items: z.array(scheduleSchema) }).strict();
 const taskDataSnapshotSchema = z.object({
-  source: z.enum(["workbench_runtime", "scada_devices", "scada_readings", "peripheral_points", "peripheral_work_orders", "peripheral_observations", "peripheral_test_data"]),
+  // Snapshot sources include product MCP queries; access-request scopes remain unchanged.
+  source: z.enum(["workbench_runtime", "scada_devices", "scada_readings", "peripheral_points", "peripheral_work_orders", "peripheral_observations", "peripheral_test_data",
+    "get_analysis_run",
+    "get_element_gis_properties",
+    "get_element_properties",
+    "get_model_baseline_at_time",
+    "get_network_options",
+    "get_network_schema",
+    "get_scada_device",
+    "get_sensor_placement_run",
+    "get_source_gis_properties",
+    "get_task_context",
+    "get_task_evidence",
+    "get_workbench_runtime",
+    "list_analysis_results",
+    "list_analysis_runs",
+    "list_elements",
+    "list_scada_devices",
+    "list_sensor_placement_runs",
+    "query_analysis_series",
+    "query_analysis_values",
+    "query_element_history",
+    "query_model_baseline_field",
+    "query_model_baseline_series",
+    "query_peripheral_data",
+    "query_scada_readings",
+    "run_burst_detection",
+    "run_simulation",
+    "run_contaminant_simulation",
+    "analyze_valve_closure",
+    "analyze_flushing",
+    "analyze_pressure_regulation",
+    "optimize_sensor_placement",
+    "run_burst_location",
+    "analyze_valve_isolation",
+    "run_leakage_identification",
+  ]),
   source_ref: z.string().min(1),
   queried_at: workbenchDateTimeSchema,
   origin: z.enum(["system_data", "synthetic"]),
@@ -286,6 +333,7 @@ export type OperationalTaskEvidence = z.infer<typeof evidenceSchema>;
 export type OperationalTaskNotification = z.infer<typeof notificationSchema>;
 export type OperationalTaskReport = z.infer<typeof reportRevisionSchema>;
 export type OperationalTaskReportPage = z.infer<typeof reportPageSchema>;
+export type OperationalTaskProgress = z.infer<typeof progressSnapshotSchema>;
 export type OperationalTaskSchedule = z.infer<typeof scheduleSchema>;
 export type OperationalTaskDataSnapshot = z.infer<typeof taskDataSnapshotSchema>;
 export type ExternalWorkOrderSnapshot = z.infer<typeof externalWorkOrderSnapshotSchema>;
@@ -342,6 +390,12 @@ export async function fetchOperationalTaskReports(taskId: string, offset: number
   ));
 }
 
+export async function fetchOperationalTaskPresentationCursor(taskId: string, getAccessToken?: AccessTokenProvider, signal?: AbortSignal) {
+  return investigationPresentationCursorSchema.parse(await operationalTaskRequest(
+    `/operational-tasks/${encodeURIComponent(taskId)}/presentation-cursor`, getAccessToken, { signal }
+  ));
+}
+
 export async function createOperationalTask(input: CreateOperationalTaskInput, getAccessToken?: AccessTokenProvider) {
   return operationalTaskSchema.parse(await operationalTaskRequest(
     "/operational-tasks",