Make test results time-invariant

[ojustino]

Progress toward figuring out a way to reliably reverse adjustments made to historical price data (splits, dividends, etc.) so test results remain consistent over time. At the moment, results for a script I run today will be slightly different in a few months and noticeably different in a few years.

Comparisons of strategies against each other calculated at the same time are OK, but comparing a calculation from 2022 to one from today is not apples-to-apples. For example, maybe the differently-adjusted price data leads a strategy to sell an extra share of an asset during a rebalance in today's simulation where it wouldn't have done so in the past. The differences aren't large, but it would be nice to achieve full reproducibility.

[ojustino]

A comment in test_suite.py suggests that run_time_loop() may double count dividends? Investigate again

[ojustino]

I think the latest commits finally reflect an approach that will lead to consistent test results over time, along with some quality of life additions (AttrDict to treat the assets dictionaries and their dataFrames as class attributes, accepting string dates in the simulator in addition to pandas Timestamps and datetimes, more clarity with the cash_out_dividends argument).

As a result, simulations can now be made time-invariant, so I will merge to avoid bloating the PR. There is more to investigate with consistency between dividend-adjusted and non-dividend-adjusted results given equivalent portfolios over identical time windows. I hope to modify the tests' assert statements after that's been diagnosed since a fix would produce different numbers than I'd calculate now anyway.