test: `FiscalYearGroupedSplitter.split` never exercises the no-prior-training-data skip branch

[shivamlalakiya]

What's wrong

FiscalYearGroupedSplitter.split skips a fold with continue when no training
row exists strictly before test_fy - gap_years. Every fixture in
tests/test_model_selection.py has enough fiscal years of history that this
never fires, so the skip path is untested — a splitter silently dropping a fold
short of n_splits is exactly the kind of thing that should be pinned with a
regression test.

Where

File	Locate with
philanthropy/model_selection/_temporal_donor_splitter.py	grep -n "No training data before this test year" philanthropy/model_selection/_temporal_donor_splitter.py

What to change

No source change — this is test-only.

1. Open tests/test_model_selection.py and look at the existing split()
   fixtures (locate with grep -n "def test_" tests/test_model_selection.py).
2. Add a test with fiscal years arranged so the earliest requested test year
   has no fiscal year strictly before it once gap_years is applied (e.g.
   groups starting at the very first fiscal year present, with n_splits
   covering it). Assert the splitter yields fewer than n_splits folds instead
   of raising or yielding an empty training set.

Tests to add or extend

• File: tests/test_model_selection.py
• Add: test_split_skips_fold_with_no_prior_training_data

Done when

python -m pytest tests/test_model_selection.py --cov=philanthropy.model_selection --cov-report=term-missing -q

philanthropy/model_selection/_temporal_donor_splitter.py reads 100% covered
with an empty Missing column.