fix(tui): repair layout overflow + streaming dup; polish UI/state

Two real bugs squashed:
- Layout: giving the header box and scrollbox an explicit width in a
  column-flex root made OpenTUI overflow vertically, pushing the input
  row and status bar entirely off-screen (only header+one message showed).
  Root cause confirmed in a minimal probe; fix = drop the redundant
  width so column children size off the root box.
- Streaming: 9router's OpenAI-compatible layer emits FULL-TEXT chunks
  (the whole transcript so far) instead of incremental deltas, so naive
  content+=token duplicated text ("Hello! 👋 How canHello! 👋 How can...").
  Added reconcileStream() that accepts both incremental and full-text
  modes (shorter-replay → ignore, growing-authoritative → adopt, else
  append), unit-tested for both.

UI/state polish:
- Rich transcript: per-role colour + labelled blocks (YOU/AI/SYS/TOOL),
  word-wrapped to terminal width, reasoning dimmed, no misleading
  "(empty)" on tool-calling assistant messages.
- Header: app · real model · session, spread across full width.
- Status bar: ● running/idle, live token usage (in/out), autocomplete
  hint, active toast.
- Bordered autocomplete dropdown + full-width help/overlay panels.
- Turn runner streams all TurnEvents into state (tool markers, agent
  progress, workflow updates, toasts, usage) — ui.tsx is now display-only.

src/interfaces/tui/state.ts

@@ -357,15 +357,46 @@ export class TranscriptCache {
     const msg = this.messages[this.messages.length - 1];
     if (msg && msg.role === Roles.Assistant) {
       if (isReasoning) {
-        msg.reasoning += text;
+        msg.reasoning = reconcileStream(msg.reasoning, text);
       } else {
-        msg.content += text;
+        msg.content = reconcileStream(msg.content, text);
       }
       this.dirty = true;
     }
   }
 }
 
+/**
+ * Reconcile an incoming stream chunk against the accumulated text.
+ *
+ * Streaming backends disagree on chunk semantics:
+ *  - Native OpenAI/Anthropic send small *incremental* deltas.
+ *  - Some gateways/proxies (e.g. 9router's OpenAI-compatible layer) emit
+ *    *full-text* chunks, where each chunk is the complete transcript so far.
+ *
+ * Blindly concatenating full-text chunks duplicates content (partial + full),
+ * which manifests as garbled messages. This helper accepts both modes:
+ *   - a chunk that is a strict prefix of the tail → a shorter replay → ignore;
+ *   - a chunk that extends/replays the whole accumulated text → take it as the
+ *     new authoritative tail (full-text mode);
+ *   - otherwise → treat as an incremental delta and append.
+ */
+export function reconcileStream(accumulated: string, chunk: string): string {
+  if (chunk.length === 0) return accumulated;
+  if (accumulated.length === 0) return chunk;
+  if (accumulated.length >= chunk.length && accumulated.startsWith(chunk)) {
+    // Shorter (or equal) replay of what we already hold — ignore.
+    return accumulated;
+  }
+  if (chunk.length >= accumulated.length && chunk.startsWith(accumulated)) {
+    // Full-text mode: the chunk is the whole transcript so far (matching our
+    // accumulated text plus more). Adopt it wholesale.
+    return chunk;
+  }
+  // Incremental delta mode: a new suffix. Append.
+  return accumulated + chunk;
+}
+
 /* ── AppStateRest ─────────────────────────────────────────────────── */
 
 /**
@@ -403,6 +434,8 @@ export interface AppStateRest {
   turnRunner: TurnRunner | null;
   /** Abort controller for the in-flight turn (driven by AbortTurn). */
   turnAbort: AbortController | null;
+  /** Token usage from the most recent turn (surfaced in the status bar). */
+  lastUsage: { tokensIn: number; tokensOut: number } | null;
 }
 
 export const DEFAULT_HELP_TEXT = ` Zesdex TUI — Keyboard Shortcuts
@@ -474,6 +507,7 @@ export function createTuiState(
     pendingSubmit: null,
     turnRunner: null,
     turnAbort: null,
+    lastUsage: null,
   };
 }