Normalization edge case

[emitra17]

Using the normalization key to map a normalization type to an exp file creates a confusing situation when there is also a prop file with the same suffix.

Looks like what currently happens is if you use the syntax normalization = init : data.exp, all columns of simulation suffix data get normalized for all uses. So if you also had a data.prop for something not even appearing in data.exp, it would get normalized.

[wshlavacek]

Fixed: Normalization no longer affects .prop constraint columns

The problem

When normalization = init : data.exp was specified, the suffix data was extracted and normalization was applied to all columns of the simulation data with that suffix. If a data.prop constraint file shared the same suffix, its columns were also normalized — even though those columns might not appear in data.exp at all. This caused constraints to evaluate against incorrectly normalized values.

The fix

During config postprocessing (_postprocess_normalization), normalization specifications are now converted from a blanket type (e.g. 'init') to column-specific form, restricted to only the columns present in the corresponding .exp file. This applies to both:

• Per-file normalization (normalization = init : data.exp) — converted to [('init', [<exp columns>])]
• Global normalization (normalization = init) — expanded into a per-suffix dict with column-specific entries for each .exp file

Simulation columns that only appear in .prop constraint files are left untouched.

Files modified

• pybnf/config.py — restrict normalization to .exp file columns in _postprocess_normalization
• tests/test_data_class.py — two new tests verifying column-specific normalization leaves other columns untouched
• tests/test_config_class.py — updated existing assertion to match new column-specific format