Respect boolean masks when assigning to L

[ktz03]

Fixes #947, reported by @j45856021-dev.

Assigning through a boolean mask to a list-backed L treats False and True as integer indices 0 and 1. For example, values = L(10,20,30); values[[False,False,True]] = 99 produces [99,99,30] instead of [10,20,99], and an all-false mask still changes data.

Normalize Python boolean index lists with the existing mask2idxs helper before assignment. Preserve the original path for native array selectors. Update the owning notebook and regenerate fastcore/foundation.py; add scalar/iterable and all-false-mask regressions plus native NumPy selector controls.

Validation on Windows with Python 3.12.14, fastcore 2.2.34 and nbdev 3.3.25:

• nbdev-test nbs/02_foundation.ipynb --n-workers 1 --verbose: identical added tests fail on the unchanged baseline and pass with the fix.
• A 46-configuration public-API matrix covers list, tuple, L and iterator masks, scalar/iterable replacements, all-false/all-true/empty selectors and integer/slice controls. All 28 affected cases are corrected; all 18 previously working cases retain their outputs.
• nbdev-export --path nbs/02_foundation.ipynb and the official notebook-cleaning check pass.
• The full nbdev-test --n-workers 1 --verbose command was run on both versions. It retains five pre-existing failing notebooks on Windows: 01_basics, 03_xtras, 03a_parallel, 03b_net and 12_tools; the added foundation regression is the sixth failure on the baseline and passes after the fix. The full suite is not claimed as passing. No tests or CI settings were changed to suppress these failures.

This addresses Python boolean masks. NumPy boolean scalars inside index lists remain unsupported with NumPy 2.5.4. Other operating systems, documentation rendering and performance were not tested.