stat summary: add combined functions: fn_data: mean_sdl / mean_se

Stat summary can't produce the very common mean ± SD / SE chart (error bar / pointrange / ribbon). Its fn / fn_min / fn_max parameters take only point aggregations — `count, sum, mean, median, min, max, lq, mq, uq` — none of which express a spread-based interval.

What's missing is any spread-based bound (SD, SE). It's structurally different: ymin = mean − k·SD is not a single-column aggregation — it's derived jointly from the center and the dispersion of the same column, so it can't be modeled as an independent fn_min / fn_max, and there's currently no slot for it.

*Proposed*
Add a combined summary function parameter — the analog of ggplot2's fun.data — that returns center and bounds together as the computed variables ..y.., ..ymin.., ..ymax..:

* `fn_data="mean_sdl"` — mean ± k·SD (ggplot2 default k = 2)
* `fn_data="mean_se"` — mean ± k·SE

with a multiplier parameter (ggplot2 passes fun.args=list(mult=…)), e.g. `fn_args={'mult': 1}`. This is a new mechanism alongside fn / fn_min / fn_max (use one or the other), mirroring ggplot2's fun vs fun.data. Ideally `stat="summary"` + `fn_data` sets the default ymin/ymax aesthetics so they needn't be re-mapped.

*Desirable API*

```Python
# cleanest — the stat provides default y/ymin/ymax
geom_pointrange(data=df, stat="summary", fn_data="mean_sdl") # mean ± 2·SD
geom_pointrange(data=df, stat="summary", fn_data="mean_se", fn_args={'mult': 1})

# explicit form (also valid, for control)
geom_pointrange(data=df, stat="summary", fn_data="mean_sdl",
mapping=aes(y='..y..', ymin='..ymin..', ymax='..ymax..'))
```

Start at the stat="summary" implementation and trace how fn, fn_min, and fn_max are handled, then inspect the existing summary tests if available. Define the combined fn_data and fn_args behavior for mean_sdl and mean_se, including default y, ymin, and ymax aesthetics; done means both proposed examples produce the expected center and bounds.

Written by the indexing model from the issue text.