fix: handle read-only arrays in DunedinPACE quantile normalization (#195)

[IsraelAfangideh]

Summary

Fixes #195.

DunedinPACE.predict raises ValueError: assignment destination is read-only during quantile normalization:

File ".../biolearn/dunedin_pace.py", line 61, in quantile_normalize_using_target
    column_data[has_decimal_above_0_4] = 0.5 * (
ValueError: assignment destination is read-only

Root cause

quantile_normalize_using_target mutates its input array in place. The array is passed in as dataframe_transposed.values. In pandas >= 3.0 (and with newer numpy), DataFrame.values returns a read-only array by default (copy-on-write), so the in-place writes fail.

This is why the reporter saw the same data work under an older Python/pandas stack but break on a newer one — it was never the Python version itself, but the pandas/numpy versions that ship with it.

Fix

Operate on a writable float copy at the top of the function:

data = np.array(data, dtype=float)

This resolves the crash and, as a bonus, stops the function from mutating the caller's array (it now returns a fresh array as the name implies).

Testing

• Reproduced the exact ValueError on the default pandas 3.0 path before the change.
• Added regression tests in biolearn/test/test_dunedin_pace.py covering: read-only input, no mutation of the caller's array, and correct mapping onto the target distribution.
• Verified end-to-end against test_model.py::test_dunedin_pace_normalization: on master it fails at the read-only line; with this change it gets past normalization.

Note

While verifying end-to-end on a pandas 3.0 environment, I noticed a separate, pre-existing failure further downstream in the DunedinPACE test path (a KeyError from an integer RangeIndex where CpG labels are expected). It is unrelated to this read-only fix and is masked on master because the read-only error aborts earlier. I've kept it out of this PR and am happy to open a separate issue/PR for it.

[sarudak]

I think this comment could be condensed significantly

[IsraelAfangideh]

Good call — condensed it to two lines, keeping just the why (read-only DataFrame.values under pandas 3.0 copy-on-write breaking the in-place writes). Pushed in bf51b76. Thanks for the review!