Improve generator quality and fix data gaps (Apr-May 2026)

Data:
- Add 12 missing draws (2026-01-09, 2026-04-03 to 2026-05-15)
- Retrain LSTM and RF models on complete 952-draw dataset
- Run retroactive learning for all 12 skipped draws

Automation:
- Add eurojackpot-zahlen.eu web scraper as fallback when APIs fail
- Fixes recurring gap problem caused by Sazka detail-endpoint change

Generator improvements:
- Add hard structural filters (sum 95-165, min 1 even + 1 odd per tip)
- Add long-term frequency prior as learner dampener (alpha=0.2)
- Fix ENSEMBLE strategy generating duplicate tips (greedy → weighted random)
- Fix non-vectorized frequency calculation (stack().value_counts())
- Fix Division-by-Zero edge case in set_frequency_prior
- Fix potential KeyError on euro_ai_score in structural filter loop

.gitignore:
- Exclude data/backups/, weekly tip CSVs, performance reports

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitignore

@@ -27,3 +27,9 @@ logs/
 
 # Results (optional - uncomment if you don't want to track results)
 # results/
+
+# Backups und generierte Outputs
+data/backups/
+data/*.backup*
+data/generated_tips/weekly_tips_*.csv
+data/performance_reports/