feat: add plot_dist for posterior density plots

[RussellSB]

Summary

• Add plot_dist() for posterior density plots on PathModel and the three draw-backed result objects: EffectResult, EstimandResult, and DoResult.
• Resolve names in one call, whether labeled coefficients ("a"), defined parameters ("indirect"), or raw posterior variables ("sigma_Y"); coords= disambiguates multi-coordinate variables.
• Add a public plot_density() renderer in pathmc.plotting (KDE curve + translucent fill + optional dashed reference line).

Issue links

• Implements Add convenient plot functions for quicker path to visualisation #323

Examples

Fitted mediation model (X -> M -> Y with a direct effect):

spec = "M ~ a*X\nY ~ b*M + c*X\nindirect := a*b\ntotal := c + a*b"
model_med = pathmc.model(spec, data=df)
model_med.fit()

1. Model: labeled coefficients

1.1 Plain

model_med.plot_dist(var=["a", "b", "c"])

1.2 Reference lines at the true values

model_med.plot_dist(var=["a", "b", "c"], ref=[0.5, 0.8, 0.3])

1.3 Plus custom colors

model_med.plot_dist(
    var=["a", "b", "c"],
    ref=[0.5, 0.8, 0.3],
    color=["lightblue", "lightcoral", "lightgreen"],
)

2. Model: defined parameters

model_med.plot_dist(var=["indirect", "total"], ref=[0.4, 0.7])

3. EffectResult: path effect

effect = model_med.effect("X -> M -> Y")
effect.plot_dist(ref=0.4)

4. EstimandResult: ATE

ate = model_med.ate("Y", "X", values=(0.0, 1.0))
ate.plot_dist(ref=0.65)

5. DoResult: interventional outcome

do_result = model_med.do(set={"X": 1})
do_result.plot_dist(var="Y")

Design decisions

• var / ref naming: generalised from the issue's var_names / ref_val so one name works across all four entry points.
• color optional: str | list[str] on the model; a single str on result objects.
• plot_dist() returns None (calls plt.show() internally).
• Shared axes + legend, no titles: multiple densities overlay on one plot.
• plot_density() is public: exported from pathmc.

Questions

• Is it OK that I made plot_density() public? It surfaces in the API reference and is directly usable for custom cases (comparing a posterior to a ground-truth distribution, for instance).
• What would be your thoughts on a plot_params argument for figure-level knobs (figsize, title, axis labels)? A future idea to extend on this method's customisation. Deferred and not blocking; happy to follow up in a later PR if there's appetite.

Test plan

• uv run pytest tests/test_plot_dist.py -x -v
• make test-fast
• make lint

Made with OpenCode Go (and a moderate amount of AI nagging)