Return the axis from AalenJohansenFitter.plot_cumulative_density

[VenishPaneliya]

AalenJohansenFitter.plot_cumulative_density returns None

The method documents:

        Returns
        -------
        ax:
            a pyplot axis object

but the body ends with a bare call:

        if not self._calc_var:
            kwargs["ci_show"] = False
        _plot_estimate(self, estimate=self._estimate_name, **kwargs)

Every sibling plot method returns that call — return _plot_estimate(self, estimate=..., **kwargs) — so this is the only one that drops the axis:

>>> type(KaplanMeierFitter().fit(T, E).plot_cumulative_density())
matplotlib.axes._axes.Axes
>>> type(AalenJohansenFitter(calculate_variance=False).fit(T, E, event_of_interest=1).plot_cumulative_density())
NoneType

so ax = ajf.plot_cumulative_density(); ax.set_title(...) raises AttributeError. Added the return.

Regression test: TestAalenJohansenFitter::test_plot_cumulative_density_returns_axis. It fails on master and passes with the fix. I put it in test_estimation.py rather than test_plotting.py deliberately — the TestPlotting class is skipif("DISPLAY" not in os.environ), so a test there wouldn't run in headless CI.

Two documented defaults in CoxTimeVaryingFitter

• fit(robust=False) was documented as (default: True). Every other fitter documents robust: bool, optional (default=False), so this one was the outlier — and it's a meaningful one, since it told users the Huber sandwich (Wei–Lin) robust standard errors were on by default when they aren't. I matched the house wording.
• _newton_raphson_for_efron_model(show_progress=False) was documented as (default: True).

I left BreslowFlemingHarringtonFitter.fit's alpha: float, optional (default=0.05) alone even though the signature is alpha=None — that wording is used for alpha in 20 places across the fitters to describe the effective default, so it's a convention rather than a mistake.

Tests

TestAalenJohansenFitter and TestCoxTimeVaryingFitter together: 45 passed with the change against 44 on master — the difference is the new test.

black --line-length 130 --check reports the three touched files as needing reformatting, but it does so identically on an unmodified checkout, so I didn't reformat unrelated code.