Add failure-path tests for data download and parquet cache

[Copilot]

Description

StockDataDownloader had no tests for failure scenarios and no production-side handling for corrupted or schema-drifted parquet cache files, leaving the pipeline vulnerable to silent crashes on common real-world failures.

Changes in this pull request

Production (src/data/stock_data_downloader.py)

• Add module-level REQUIRED_COLUMNS: frozenset — the five price columns yfinance always returns
• Extract _load_cached_parquet(path, symbol) -> (df | None, needs_refresh) static helper that consolidates all cache validation:
  • Wraps pd.read_parquet in a try/except — corrupted/truncated files trigger re-download instead of propagating an exception
  • Checks for schema drift (REQUIRED_COLUMNS - set(df.columns)) — missing columns trigger re-download
  • Staleness check (existing behaviour, now co-located)
• _download_single now delegates to the helper, eliminating the df = None guard pattern
• Fix ty typecheck errors: add cast to the typing import; add assert df is not None in the cached path to narrow pd.DataFrame | None before the dict assignment; use cast(pd.DataFrame, ticker.history(...)) and assert isinstance(df, pd.DataFrame) after the retry loop to narrow the union type produced by yfinance's untyped return

Tests (tests/data/test_stock_data_downloader.py)

• TestDownloadFailurePaths — new class covering all four production failure modes:
  • test_partial_failure_returns_successful_data — @pytest.mark.parametrize over (1/3, 2/4, 3/3) failing tickers; asserts surviving tickers are returned and failures are tracked in failed
  • test_all_tickers_fail_returns_empty_dict — all tickers raise; asserts result == {} with no crash
  • test_corrupted_parquet_triggers_redownload — garbage bytes written to cache path; asserts yfinance is called and symbol is in result
  • test_schema_drift_triggers_redownload — cache missing Close/Volume; asserts re-download fires and full schema is present in result

[ch55secake]

@copilot typecheck failing

[Copilot]

@copilot typecheck failing

Fixed in ddddd15. Three changes to stock_data_downloader.py:

1. Added cast to the typing import
2. assert df is not None in the cached path (before the dict assignment) — ty couldn't narrow from the _load_cached_parquet return type
3. cast(pd.DataFrame, ticker.history(...)) + assert isinstance(df, pd.DataFrame) after the retry loop — ty saw yfinance's untyped return as Unknown and was unioning it with the outer-scope DataFrame | None

All 36 tests pass and ty check src/ is clean.