Add workflow-status telemetry view and make panel modes on-demand

Workflow-status: a new WorkflowProgressTracker singleton taps the
suppressed dynamic-workflow subagent events (those carrying
parent_tool_use_id) into per-session telemetry without altering the
existing guard; the snapshot is serialized into the new chains.workflow_info
column on FinalizeChain (ChainId-matched so a panel member of the same
session cannot clear it) and exposed via GET /api/sessions/{id}/workflow-status
(live -> persisted -> none). The header menu gains a 'Workflow-Status' entry
with a modal that live-polls every 2s while running and self-stops on close.

Panel on-demand: Fusion/MoA is now a transient per-request multiModelMode
flag (like plan/ultracode) instead of a persisted session column. Session
settings keep only the panel members + moa_layers; the mode <select> was
removed and the composer dropdown drives the mode per request.

Docs: ARCHITECTURE.md, ARCHITECTURE_QUICK.md and README.md updated for both
features plus the already-shipped z.ai/GLM provider, GLM 5.2+ effort
passthrough, web-search-prime MCP and the ultracode toggle.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Services/ClaudeCliService.cs

@@ -21,11 +21,13 @@ public class ClaudeCliService
     private readonly string _cliPath;
     private readonly bool _isAvailable;
     private readonly ServerSettingsService _settings;
+    private readonly WorkflowProgressTracker _workflowTracker;
 
-    public ClaudeCliService(ILogger<ClaudeCliService> logger, ServerSettingsService settings)
+    public ClaudeCliService(ILogger<ClaudeCliService> logger, ServerSettingsService settings, WorkflowProgressTracker workflowTracker = null)
     {
         _logger = logger;
         _settings = settings;
+        _workflowTracker = workflowTracker;
         _cliPath = FindCliPath();
         _isAvailable = !string.IsNullOrEmpty(_cliPath) && File.Exists(_cliPath);
 
@@ -90,6 +92,44 @@ public class ClaudeCliService
         return major > 5 || (major == 5 && minor >= 2);
     }
 
+    /// <summary>
+    /// Greift aus einem (im Haupt-Stream unterdrückten) Subagent-Turn die Telemetrie ab:
+    /// die vom Subagenten aufgerufenen Tool-Namen und seine Token-Nutzung. Rein additiv für den
+    /// Workflow-Status – wirft nie (Telemetrie darf den Stream-Parser nicht stören).
+    /// </summary>
+    private void TapSubagentEvent(string sessionId, string chainId, string parentToolUseId, JsonElement json)
+    {
+        try
+        {
+            if (!json.TryGetProperty("message", out var msg))
+                return;
+
+            var toolNames = new List<string>();
+            if (msg.TryGetProperty("content", out var content) && content.ValueKind == JsonValueKind.Array)
+            {
+                foreach (var item in content.EnumerateArray())
+                {
+                    if (item.TryGetProperty("type", out var it) && it.GetString() == "tool_use"
+                        && item.TryGetProperty("name", out var nm))
+                    {
+                        var n = nm.GetString();
+                        if (!string.IsNullOrEmpty(n)) toolNames.Add(n);
+                    }
+                }
+            }
+
+            long inTok = 0, outTok = 0;
+            if (msg.TryGetProperty("usage", out var usage))
+            {
+                if (usage.TryGetProperty("input_tokens", out var it2) && it2.ValueKind == JsonValueKind.Number) inTok = it2.GetInt64();
+                if (usage.TryGetProperty("output_tokens", out var ot2) && ot2.ValueKind == JsonValueKind.Number) outTok = ot2.GetInt64();
+            }
+
+            _workflowTracker.RecordSubagent(sessionId, chainId, parentToolUseId, toolNames, inTok, outTok);
+        }
+        catch { /* Telemetrie darf den Stream nie stören */ }
+    }
+
     private ProcessStartInfo CreateCliStartInfo(List<string> args, bool redirectInput = false, string effort = null, string workingDirectory = null, string modelId = null)
     {
         var psi = new ProcessStartInfo
@@ -219,6 +259,8 @@ public class ClaudeCliService
         string mcpConfigPath = null,
         bool hasUnityMcp = false,
         string effort = null,
+        string sessionId = null,
+        string chainId = null,
         CancellationToken cancellationToken = default)
     {
         if (!_isAvailable)
@@ -354,7 +396,7 @@ public class ClaudeCliService
             args.Add(mcpConfigPath);
         }
 
-        return await RunCliStreamingAsync(args, prompt, onEvent, cancellationToken, effortToUse, workspace, modelId);
+        return await RunCliStreamingAsync(args, prompt, onEvent, cancellationToken, effortToUse, workspace, modelId, sessionId, chainId);
     }
 
     private async Task<CliChatResult> RunCliStreamingAsync(
@@ -364,7 +406,9 @@ public class ClaudeCliService
         CancellationToken cancellationToken,
         string effort = null,
         string workspace = null,
-        string modelId = null)
+        string modelId = null,
+        string sessionId = null,
+        string chainId = null)
     {
         var result = new CliChatResult();
         var fullOutput = new StringBuilder();
@@ -509,6 +553,12 @@ public class ClaudeCliService
                         && parentToolUseId.ValueKind == JsonValueKind.String
                         && !string.IsNullOrEmpty(parentToolUseId.GetString()))
                     {
+                        // Subagent-Event: weiterhin aus dem Haupt-Stream AUSBLENDEN (continue) –
+                        // hier wird nur die Telemetrie für den Workflow-Status ABGEGRIFFEN
+                        // (Tool-Aufrufe + Subagent-Tokens), ohne den sichtbaren Verlauf zu verändern.
+                        // Die konsolidierten Subagent-Turns kommen als type:"assistant" (mit usage).
+                        if (_workflowTracker != null && !string.IsNullOrEmpty(sessionId) && type == "assistant")
+                            TapSubagentEvent(sessionId, chainId, parentToolUseId.GetString(), json);
                         continue;
                     }