Massive float-comparison fix effort (v2.4)
So many bugs were found in the making of this commit.
- Dropping of duplicates in spat_df and con_spat_df failed if floats were off by machine precision. Now has rounding when checking for dupes.
- Changed std function in sig_models from pandas (unbiased) to numpy (biased) estimator
- Changed old, weird-ass selection logic in sig_models and spec_models to select based on k and not by colnames
- Changed float comparison in data_spectrum to use np.isclose instead of ==
- Updated tests to catch the float-comparison issue for dropping duplicates and spec_models
- Changed version to 2.4.0dev