Improve test data

[proway2]

Testdata was created with FORTRAN, some values are coarse, especially all SV values. Allowed error (used in test cases) for this values is very high.

• Check if it's possible to create more accurate testdata with C version.
• Generate the new testdata
• Implement a way to digest the new testdata and automatically create tests bases on it.
• Refactor tests - decrease max allowed error for SV fields.

[proway2]

If you request the C implementation to produce data for an interval of dates, it doesn't produce SV values, unless for the last value. Although C implementation produce more accurate values (fraction part is present for all values).

[proway2]

C implementation produces NaN for near true pole values (close to 90.0 and -90.0) and for magnetic poles 86.49N 162.87E for 2020.0. It's impossible to use values produced by C version and at the same time have near poles values calculated.
There is a way of mixing values produced by C and values produced by FORTRAN (near pole values), but it's unclear what it really brings. Declination and inclination (produced by FORTRAN) already tested with very high accuracy.