Improve Video Learning Agent with action-based detection and analysis toggle

- Change video detection from screenshot hash to action-based (Swipe detection)
- Add enable_analysis toggle to disable VLM screenshot analysis
- Improve task prompt to prevent VLM from stopping prematurely
- Add debug logging for action detection troubleshooting
- Fix ModelResponse attribute error (content -> raw_content)

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

dashboard/static/js/video-learning.js

@@ -19,6 +19,7 @@ const VideoLearningModule = {
             targetCount = 10,
             category = null,
             watchDuration = 3.0,
+            enableAnalysis = true,
         } = options;
 
         try {
@@ -28,6 +29,7 @@ const VideoLearningModule = {
                 target_count: targetCount,
                 category: category,
                 watch_duration: watchDuration,
+                enable_analysis: enableAnalysis,
             });
 
             this.currentSessionId = response.data.session_id;