Float formatter learn_rounding_scheme doesn't work on all digits

[npatki]

Environment Details

• RDT version: 1.2.1

Error Description

When using a FloatFormatter with learn_rounding_scheme=True, we expect the transformer to learn the maximum # of significant digits. In practice, we see the following:

1. If data has 0-14 digits, then the transformer learns rounding scheme [Working as intended]
2. If the data has 15+ digits, then the transformer learns 0 digits, producing whole numbers instead [Bug]

We expect that case 2 to work. Or as a fallback, at least stop enforcing the rounding if there are already a large number of digits.

Steps to reproduce

import pandas as pd
from rdt import HyperTransformer
from rdt.transformers.numerical import FloatFormatter

# create test data with 16 digits
test_data = pd.DataFrame(data={
    'column': [1.1234567890123456]
})

ht = HyperTransformer()
ht.set_config({
    'sdtypes': { 'column': 'numerical' },
    'transformers': { 'column': FloatFormatter(learn_rounding_scheme=True)}  
})

t = ht.fit_transform(test_data)
ht.reverse_transform(t)

Output: 1.0 (no digits learned)