Price Signals Dispatch runs slightly differently on Windows and Linux

[brtietz]

This is most obvious in the test CMPvsamv1BatteryIntegration_cmod_pvsamv1.ResidentialDCBatteryModelPriceSignalDispatch, where Windows has an average cycle depth 0.4 percentage points higher than Linux, hence the high tolerance.

A suspicious candidate is the sorting algorithms, as these can be different in different implementations of the standard library. Clean up and reduce the tolerances for the tests.

[brtietz]

Closed in 9cf27a1

[brtietz]

This issue recurred in #675. See that issue for some additional debugging details. The timestep numbers may drift as defaults or PVSamV1 changes. The general approach is to look for when battery SOC diverges, and then find the dispatch step that causes the divergence.