fix(engine): raise output-token + step ceilings (mid-response cutoff)

User repro: V4 thinking on hard prompts (~107s of thinking) randomly
"stops mid-response", more often when starting in Agent mode and
switching to YOLO. Two ceilings were too tight:

1. TURN_MAX_OUTPUT_TOKENS: 4096 → 32768
   `reasoning_content` from V4 thinking can easily exceed 4K tokens on
   hard problems. Once the per-turn output budget exhausts, the API
   closes the SSE stream with `finish_reason: "length"` and the visible
   reply ends up empty — surfaced as the assistant "stopping randomly".
   32K leaves comfortable headroom for thinking + the visible reply on
   every realistic turn while staying well below DeepSeek V4's
   1M-context output ceiling.

2. max_steps: 100 → u32::MAX (effectively unlimited)
   100 was hitting the ceiling on long multi-step plans (wide refactors,
   sub-agent orchestration) and presenting as the agent "giving up
   mid-task". V4's 1M context window means there's no good reason to
   cap steps administratively. Users can still interrupt with Ctrl+C /
   Esc; a turn naturally ends when the model stops emitting tool calls.

All 54 turn tests pass; full workspace clippy + fmt remain clean.

crates/tui/src/core/engine.rs

@@ -263,8 +263,12 @@ const STREAM_CHUNK_TIMEOUT_SECS: u64 = 90;
 const STREAM_MAX_CONTENT_BYTES: usize = 10 * 1024 * 1024; // 10 MB
 /// Maximum wall-clock duration for a single streaming response.
 const STREAM_MAX_DURATION_SECS: u64 = 300; // 5 minutes
-/// Max output tokens requested for normal agent turns.
-const TURN_MAX_OUTPUT_TOKENS: u32 = 4096;
+/// Max output tokens requested for normal agent turns. Bumped from 4096 to
+/// 32768: V4 thinking models can consume 8-15K reasoning tokens on hard
+/// prompts; the old 4K ceiling exhausted the budget, the API closed the
+/// SSE stream with `finish_reason: "length"`, and the visible reply ended
+/// up empty (surfaced as the assistant "stopping mid-response").
+const TURN_MAX_OUTPUT_TOKENS: u32 = 32768;
 /// Keep this many most recent messages when emergency trimming is required.
 const MIN_RECENT_MESSAGES_TO_KEEP: usize = 4;
 /// Allow a few emergency recovery attempts before failing the turn.

crates/tui/src/core/engine/tests.rs

@@ -262,7 +262,7 @@ fn detects_context_length_errors_from_provider_payloads() {
 fn context_budget_reserves_output_and_headroom() {
     let budget = context_input_budget("deepseek-v3.2-128k", TURN_MAX_OUTPUT_TOKENS)
         .expect("deepseek models should have known context window");
-    let expected = 128_000usize - 4_096usize - 1_024usize;
+    let expected = 128_000usize - (TURN_MAX_OUTPUT_TOKENS as usize) - 1_024usize;
     assert_eq!(budget, expected);
 }
 

crates/tui/src/tui/ui.rs

@@ -314,7 +314,15 @@ fn build_engine_config(app: &App, config: &Config) -> EngineConfig {
         trust_mode: app.trust_mode,
         notes_path: config.notes_path(),
         mcp_config_path: config.mcp_config_path(),
-        max_steps: 100,
+        // Effectively unlimited. V4 has a 1M context window and the user
+        // wants the model running until it's actually done. The previous cap
+        // of 100 hit the ceiling on long multi-step plans (wide refactors,
+        // sub-agent orchestration) and presented as the agent "giving up
+        // mid-task". `u32::MAX` is the type ceiling; users can still
+        // interrupt with Ctrl+C / Esc, and a turn naturally ends when the
+        // model stops emitting tool calls. A real runaway is rare and
+        // human-noticeable; we trust the operator over a hard step cap.
+        max_steps: u32::MAX,
         max_subagents: app.max_subagents,
         features: config.features(),
         compaction: app.compaction_config(),