LLM: streaming + rich context + 32×32 frames + proxy log

- Streaming: requestSongTweak now uses streamObject when a progress
  callback is provided (via handleLlmTweak), emitting partial {rationale,
  edits} and updating toast/console as text arrives; falls back to
  generateObject if gateway doesn't support streaming. Models always
  called via provider.chat(modelId) / /v1/chat/completions, the correct
  path for meta/muse-spark-1.3-contributor via gateway.

- Rich context: spec.js gains FULL control wording (any amount/arrangement/
  bpm allowed), and detailed payload — trackSummary with bandBalance,
  tempo, normalization (per-feature lo/hi), sectionsDetailed (energy/
  flux/centroid/flatness/width/slope/endEnergy/duration), stats
  (kinds, avg/max energy, bars). LLM told You have FULL control.

- Low-res vision: new src/llm/thumbs.js — captureThumbnails(show, {8,32})
  samples section midpoints, warms up state, renders via Show, box-averages
  down to 32×32 JPEG data URLs, restores timeline. main.js now captures 8
  frames with two animation frames of pause, then passes them as {type:
  image, image: dataUrl} parts in the user message (llmTweak.js
  attachFramesToMessages).

- Proxy logging (no bearer): vite.config.js llmGatewayProxy now logs
  sanitized requests/responses to logs/llm-proxy.log (JSON lines, 1MB
  rotation, redacts apiKey/authorization from path/body/headers) so the
  coding agent can inspect prompts without re-calling. .gitignore adds
  logs/.

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

flow-state/src/main.js

@@ -16,6 +16,7 @@ import { synthesizeSong } from './audio/synth.js';
 import { FeatureTrack } from './audio/FeatureTrack.js';
 import { generateLook } from './look/LookGenerator.js';
 import { SONGS } from './audio/songbank.js';
+import { captureThumbnails } from './llm/thumbs.js';
 
 const QUALITY = {
     draft: 0.5,     // half resolution — for scrubbing heavy stacks
@@ -1053,8 +1054,9 @@ async function handleLlmTweak() {
     const btn = document.getElementById('btn-llm-tweak');
     const prevLabel = btn ? btn.textContent : '';
     if (btn) btn.textContent = '… tuning';
-    // Immediate feedback — budget had a one-frame lag before, so show progress explicitly.
-    showToast(`Contacting LLM (${cfg.model})…`, { ok: true });
+    // Live progress: start capturing thumbs, then stream. Toast updates as stream arrives.
+    let llmProgressToast = `Contacting LLM (${cfg.model})…`;
+    showToast(llmProgressToast, { ok: true });
     console.info('[llm] handleLlmTweak start', {
         model: cfg.model, gateway: cfg.gatewayUrl,
         remaining, hint: hint || '(none)',
@@ -1062,13 +1064,38 @@ async function handleLlmTweak() {
         lookDirector: lookBefore && lookBefore.director,
     });
 
+    // Capture 32×32 thumbnails — low-res strip for the vision model, non-blocking.
+    showToast('Capturing thumbnails…', { ok: true });
+    let llmFrames = [];
+    try {
+        // Keep the preview responsive — allow one paint before capture.
+        await new Promise((r) => requestAnimationFrame(() => requestAnimationFrame(r)));
+        llmFrames = await captureThumbnails(state.show, { count: 8, size: 32 });
+        console.info('[llm] captured frames', llmFrames.length, llmFrames.map((f) => `${f.kind}@${f.time.toFixed(1)}s`));
+    } catch (e) {
+        console.warn('[llm] captureThumbnails failed', e);
+    }
+    showToast(`Contacting LLM (${cfg.model}) · ${llmFrames.length ? llmFrames.length + ' frames' : 'no frames'}…`, { ok: true });
+
     try {
         const result = await requestSongTweak({
             spec: specBefore,
             track: trackBefore,
             look: lookBefore,
             hint: hint || '',
+            frames: llmFrames,
             config: cfg,
+            onProgress: (evt) => {
+                if (!evt) return;
+                const text = evt.text ? evt.text.slice(0, 80) : '';
+                if (evt.stage === 'partial') {
+                    const partialKeys = evt.partial && evt.partial.edits ? Object.keys(evt.partial.edits).filter((k) => evt.partial.edits[k] != null) : [];
+                    showToast(`LLM streaming… ${text}${partialKeys.length ? ` · ${partialKeys.join(', ')}` : ''}`, { ok: true });
+                    console.info('[llm] stream', evt);
+                } else if (evt.stage === 'streaming') {
+                    showToast(`LLM streaming… ${text}`, { ok: true });
+                }
+            },
         });
 
         // Log what came back before any synthesis — this is the audit trail.