[OPEN_DEFCON_1d] Fix moving covariance estimator's estimation subsample

[sparshsah]

What this SHOULD DO at time t is

• Estimate mean over [t - horizon + 1 : t]
• Subtract that constant from each observation in [t - horizon + 1 : t]
• Take the root-mean-square of that timeseries of deviations from the constant mean

What this INSTEAD DOES at time t is

• Estimate mean over [t - horizon - horizon + 1 : t - horizon + 1]
• Subtract that from the single observation at t - horizon + 1
• Estimate mean over [t - horizon - horizon + 2 : t - horizon + 2]
• Subtract that from the single observation at t - horizon + 2
• ...
• THEN finally take the root-mean-square of that timeseries of deviations from the time-varying (because it looks backward over horizon timesteps at each timestep) mean!

Shit.

Refer to:
https://github.com/sparshsah/foggy-lib/blob/0648d816a910fcdea4a654265f50a5578f5afd23/util/foggy_pylib/stats/est/tsa.py#L179