fix: implement the proper sliding window method from the raw values

[iomz]

The correct sliding window will increase the size of dataset – however, this pipeline requires composite bandpass/rectification filter that is independent from the Myo's embedded FV mode.
Study of EMG signal filtering is required and will be done in a separate issue.

The further analysis should reveal the reason for why those half-baked solution can still yield an "OK" quality, but not truly reaching the comfort.

[iomz]

In addition to the real-time aggregation problem in #52, this sliding window method will add an additional parameter: delta time to slide, and the slid time windows all need to have the same number of data points to be fed into classifiers...

Given the sampling rate is not stable (although, it's not too bad) I need to come up with a careful design here. I observed the data feed was extremely once at one occasion, but that must be from other factors (I hope).

[iomz]

cf. the speedometer scripts measures the throughput of the data: https://github.com/Interactions-HSG/MyoKTROS/blob/main/scripts/speedometer.py