Fix Portfolio Optimizer failing on mixed datetime resolutions between prices and FX closes

[alfwro13]

Summary

• Portfolio Optimizer run failed with MergeError: incompatible merge keys [0] dtype('<M8[us]') and dtype('<M8[ms]') raised from fx_conversion_helpers.fx_levels_on().
• Cause: under pandas 3 the X-ray return dates (parsed from stored strings) are datetime64[us], while the FX closes read back from Parquet are datetime64[ms]; merge_asof requires identical key dtypes.
• Fix: align the FX index to the dates' resolution before the merge, building the right-hand frame without mutating the in-memory cached FX series. Same helper serves the Strategy Backtester, so its currency conversion is covered too.

Test plan

• pytest tests/test_fx_conversion_helpers.py tests/test_portfolio_optimizer_engine.py tests/test_fx_drag_engine.py tests/test_accounts_engine.py tests/test_strategy_backtest_engine.py tests/test_strategy_backtest_runs.py
• New tests fail on the previous code (13 failures) and pass with the fix: all s/ms/us/ns date/FX resolution pairs, plus an end-to-end returns_in_base case with string-parsed dates against millisecond FX