Feature: support double precision option

[drewsilcock]

This adds a keyword argument to the Comtrade() constructor called use_double_precision (default: false) which causes the time and analog data arrays to use double-precision floating point precision instead of single precision.

This also adds:

• Some very simple tests for the new option.
• A modification to a previously failing UTF-8 test.
• A GitHub Action to run the tests so that they run on all pushes and pull requests.

This arose from a need we have to have double-precision on the timestamps specifically - the time values are in nanosecond precision and single precision floating point only gets you 7-8 decimal points of precision.

As an example, if you take the 2013 sample file, the sampling rate is 1200 Hz and there are 40 data points which means the last data point has timestamp 39 / 1200 = 0.0325 but when loaded into the single-precision array it gets stored as 0.032499998807907104. The larger the index of the data point, the larger the discrepancy. We're dealing with files that have almost 400,000 data points so the precision errors are non-insignificant.

[dparrini]

Hi @drewsilcock , this is a nice feature. I just solved some old issues and this caused some conflicts with your PR.