Fix get_decimal_precisions IndexError on scientific-notation modes (#442)

[jbbqqf]

Summary

PublicData.get_decimal_precisions uses str(mode).split('.')[1] to count
decimal digits on continuous columns. When the column's mode renders in
scientific notation — e.g. '1e-06' for floats with magnitude <= 1e-5 or
'1e+16' for very large integer-valued floats — the string has no '.',
so split('.')[1] IndexErrors and the dice_ml.Data(...) constructor
blows up before the user can do anything. As @joeranbosma reports in #442,
this hits real datasets with currency/probability/nano-second magnitudes.

Why

Two changes:

1. Fix: extract _decimal_precision_of(value) static helper that
   preserves the historical behaviour for plain decimal reprs ('0.25' →
   2) and handles scientific-notation reprs by combining the mantissa's
   decimal count with the exponent ('1e-06' → 6, '2.5e-3' → 4,
   '1e+16' → 0). Re-deriving from the scientific form rather than from a
   re-formatted %.20f string avoids leaking float64 ↔ float32 round-off
   junk digits into the result.

2. Regression test: parametrized TestDecimalPrecisionsScientificNotation
   covering small (1e-6), small-with-mantissa (2.5e-3), large (1e16), and
   ordinary (0.5) modes. The first and third fail on origin/main with
   the exact IndexError: list index out of range from the bug report.

Reproduce BEFORE/AFTER yourself (copy-paste)

set -e
cd /tmp && rm -rf DiCE-442 && git clone https://github.com/interpretml/DiCE.git DiCE-442
cd DiCE-442
pip install -q -e . pytest

git fetch https://github.com/jbbqqf/DiCE.git fix/442-decimal-precisions-scientific
git checkout FETCH_HEAD -- tests/test_data_interface/test_public_data_interface.py

# --- BEFORE: fix not applied (production code from origin/main) ---
git checkout origin/main -- dice_ml/data_interfaces/public_data_interface.py
python -m pytest tests/test_data_interface/test_public_data_interface.py::TestDecimalPrecisionsScientificNotation -q || echo \"BEFORE: 2 failed (IndexError) — expected\"
# Expected: 2 failed, 2 passed (IndexError on values0-6 and values2-0)

# --- AFTER: fix applied ---
git checkout FETCH_HEAD -- dice_ml/data_interfaces/public_data_interface.py
python -m pytest tests/test_data_interface/test_public_data_interface.py::TestDecimalPrecisionsScientificNotation -q
# Expected: 4 passed

What I ran locally

• pytest tests/test_data_interface/test_public_data_interface.py → 39 passed (4 new + 35 existing)
• Full test_public_data_interface suite still green
• Confirmed 2/4 of the new tests fail on origin/main with IndexError

Edge cases

input mode repr	precision returned	rationale
'0.25'	2	unchanged from historical behaviour
'1e-06'	6	mantissa has no decimals; exponent -6 ⇒ 6
'2.5e-3'	4	mantissa has 1 decimal; exponent -3 ⇒ 1 + 3 = 4
'1e+16'	0	positive exponent collapses fractional digits to 0
'17' (int-valued float)	0	new explicit branch (was implicit before)
Mixed-precision modes (e.g. [0.5, 0.25])	max across modes	unchanged

AI disclosure

This change was prepared with the assistance of Claude (Anthropic).
The author reviewed every line and is responsible for the final result.