Add true end-to-end integration tests across all training pipelines

[Copilot]

Description

Adds true end-to-end integration tests covering all three training pipeline classes and all registered trading strategies. Each pipeline is exercised through the complete sequence: read_tickers → download_data → prepare_features → train → simulate, with StockDataDownloader mocked so no network calls are made. Two pre-existing bugs in SingleModelTrainingPipeline were also fixed as a prerequisite for testing it.

Changes in this pull request

• Added tests/integration/__init__.py and tests/integration/test_full_pipeline.py with 24 end-to-end assertions
• StackedModelTrainingPipeline: tested through train → simulate → rank_and_allocate; simulate() runs all 13 registered strategies
• SingleModelTrainingPipeline: added TestSinglePipelineAlignment and TestSinglePipelineSimulate test classes; fixed two pre-existing bugs — _test_results was never stored after evaluate(), and simulate() was not implemented (leaving the class un-instantiable as an ABC)
• TimeSeriesStackedModelTrainingPipeline: added TestTSStackedPipelineAlignment and TestTSStackedPipelineSimulate test classes; uses a single ticker because TimeSeriesStacker splits on the DataFrame index and duplicate timestamps from multiple tickers cause length mismatches
• Added fast_cat_params() helper in tests/helpers.py for minimal CatBoost parameters in tests
• All 13 registered strategies (adaptive, bb_reversion, coinflip, contrarian, directional, ema_cross, hold_days, hybrid_trend_ml, momentum, sltp, sma_trend, time_stop, volatility_adjusted) are exercised in each pipeline's simulate() call
• Replaced traceback.print_exc() with logger.error(..., exc_info=True) in SingleModelTrainingPipeline to comply with project logging guidelines