fix(coding-agent): size title/commit/speech/classifier budgets for backends that ignore disableReasoning

Sizing `maxTokens` off the static `model.reasoning` catalog flag cannot
distinguish a thinking model catalogued `reasoning: false` (e.g. Qwen3
served locally via llama.cpp, whose bundled jinja chat template defaults
`enable_thinking: true`) from a model that never emits thinking. The
tight non-reasoning budget was consumed by the thinking preamble before
the useful output could be emitted, so every affected call silently
failed with `stopReason: "length"`.

Drop the `model.reasoning` conditional across every affected online call
site and always reserve the reasoning-safe budget. `maxTokens` is a hard
cap, not a target — non-thinking completions still return in the tiny
happy-path budget.

Sites fixed:
- utils/title-generator.ts       (30   -> 1024)
- utils/commit-message-generator (60   -> 1024)
- tts/speech-enhancer            (512  -> 1536)
- auto-thinking/classifier       online path (8  -> 1024); classifyLocal
                                  keeps its separate LOCAL_ANSWER_MAX_TOKENS
- session/unexpected-stop-classifier online path (16 -> 1024);
                                  classifyLocal keeps ANSWER_MAX_TOKENS

Fixes #4355

packages/coding-agent/src/utils/commit-message-generator.ts

@@ -16,8 +16,12 @@ import { concreteThinkingLevel, toReasoningEffort } from "../thinking";
 
 const COMMIT_SYSTEM_PROMPT = prompt.render(commitSystemPrompt);
 const MAX_DIFF_CHARS = 4000;
-const COMMIT_MAX_TOKENS = 60;
-const REASONING_SAFE_MAX_TOKENS = 1024;
+// Cover the "backend ignores `disableReasoning`" case unconditionally: the
+// static `model.reasoning` catalog flag can't distinguish a thinking model
+// declared `reasoning: false` (e.g. Qwen3 served locally via llama.cpp) from
+// one that never emits thinking. `maxTokens` is a hard cap — non-thinking
+// completions still return in a handful of tokens (issue #4355).
+const COMMIT_MAX_TOKENS = 1024;
 
 /** File patterns that should be excluded from commit message generation diffs. */
 const NOISE_SUFFIXES = [".lock", ".lockb", "-lock.json", "-lock.yaml"];
@@ -101,9 +105,7 @@ export async function generateCommitMessage(
 		if (!apiKey) continue;
 
 		try {
-			const maxTokens = candidate.model.reasoning
-				? Math.max(COMMIT_MAX_TOKENS, REASONING_SAFE_MAX_TOKENS)
-				: COMMIT_MAX_TOKENS;
+			const maxTokens = COMMIT_MAX_TOKENS;
 			const response = await completeSimple(
 				candidate.model,
 				{