Closes #9891.

Problem

describe / Description raised an opaque error, deep from pandas/numpy, when given a 0-row DataFrame — even though pandas.DataFrame.describe() handles the same input fine and returns a frame of NaNs. The message depended on the column dtype, hinting at two separate unhandled spots:

• numeric column → ValueError: Length of values (2) does not match length of index (1)
• categorical column → KeyError: 2

Root cause

Both live in Description.numeric (which describe computes by default):

1. df.apply(_mode) on a 0-row frame takes pandas' empty-result path, which calls _mode on an empty column; _mode returns a 2-tuple, so pandas tries to fit 2 values into the 1-length column index → ValueError.
2. For all-categorical input the numeric frame has no columns, so the Jarque–Bera result jb has no columns and jb[2]/jb[3] (skew/kurtosis) → KeyError: 2.

Fix

Short-circuit both undefined computations to NaN when there are no observations:

• when df.shape[0] == 0, set mode/mode_counts to NaN of the right length instead of calling apply;
• when there are no observations or no numeric columns, build a NaN Jarque–Bera frame with the expected four columns.

describe now returns an empty/NaN summary (nobs=0, missing=0, other stats NaN), matching pandas.describe. Non-empty inputs are completely unchanged.

Tests

Added a parametrized test_empty_rows covering 0-row numeric, categorical, and mixed frames (asserts no error, correct columns, nobs/missing = 0, NaN stats, and describe == Description(...).frame). Full test_descriptivestats.py passes (18 tests).