Skip NaN in quantile() and median()

[vahid-ahmadi]

Closes #302

Problem

quantile had no NaN handling. NaN sorts to the end of np.argsort, and its weight still counted toward the cumulative distribution, so the inverse-CDF cutoff was pushed upward:

MicroSeries([1.0, np.nan, 3.0], weights=[1, 1, 1]).median()   # -> 3.0
MicroSeries([1.0, 3.0],         weights=[1, 1]).median()      # -> 1.0

Dropping a NaN row should give the same answer as never having had it. Instead the NaN row's weight made up a third of the distribution, moving the 0.5 cutoff past the second value onto the largest one.

To be precise about the expected value: this method is documented as inverse CDF (survey::svyquantile), so the correct answer is 1.0 — an observed value — not pandas' interpolated 2.0. This PR does not change the interpolation semantics, only the NaN handling.

Change

• skipna: bool = True on quantile and median.
• With skipna=True, NaN rows are dropped in the same mask that already drops zero-weight rows (Skip zero-weight rows in MicroSeries.quantile #287), so their weight no longer enters cumsum.
• With skipna=False, NaN is returned when any value is NaN — matching mean (Fix mean() method to properly handle skipna parameter #269), var/std (Weighted std/var; warn on unweighted cov/corr fallthrough #290).
• All-NaN input returns NaN (it already hit the "no positive weight" branch; now covered by a test).

Verification

• Three new tests. On main, test_quantile_skips_nan and test_quantile_skipna_false_propagates_nan fail; all three pass with the change.
• The existing Skip zero-weight rows in MicroSeries.quantile #287 zero-weight regression still holds: MicroSeries([10, 20, 30], weights=[0, 1, 1]).quantile(0) -> 20.
• Integer dtypes unaffected (pd.isna handles them, no np.isnan cast error).
• make test — 65 passed on pandas 3.0.5 and pandas 2.3.3.

Marked as a minor bump in the changelog fragment: default median/quantile results change for data containing NaN.

Note: the lint job will be red until #307 merges (unrelated ruff 0.16 issue on main).