cd_summary(): stations sharing a long_name give indistinguishable rows

[NewGraphEnvironment]

If we do it: a cd_summary() table over several stations that share one long_name says which row is which station. If we never do: two rows read Mean discharge / Annual with different slopes and nothing to tell them apart — while cd_plot_comparison() beside the table labels the same series Mean discharge (q_site1).

Problem

series_check() tells callers that series from several sites need distinct variable names. The natural next step for streamflow is one long_name ("Mean discharge") on every station. cd_summary() (R/cd_summary.R) replaces variable with Parameter = long_name and drops variable, so those rows become indistinguishable. Measured in #93's review with q_site1 / q_site2.

#93 handled the same case in cd_plot_comparison(): a label shared by several variables gets (variable) appended, and facets key on variable + label so nothing can merge.

Proposed Solution

• In cd_summary(), append (variable) to a Parameter shared by more than one variable — the same rule as cd_plot_comparison(), ideally lifted into one internal helper both call.
• Test: two stations, one long_name → two distinct Parameter values; registered ERA5 variables unchanged (registry long_names are unique).

A second way to get indistinguishable rows: since #97 a trend table can hold both a raw-value and an anomaly trend of one series (bind_rows(cd_trend(x), cd_trend(ano)), told apart by trend_on). cd_summary() drops trend_on, so those rows differ only by Unit, and for an absolute variable such as tmean not even by that. Whatever disambiguates stations here should also cover trend_on, or cd_summary() should carry it.

Decided (plan gate, 2026-09-29): cd_summary() carries it: a Trend on column (Value/Anomaly) after Period, added only when the table holds both scales, so single-scale output keeps its shape. Chosen over always adding the column and over suffixing Parameter. The station rule landed as the shared helper label_disambiguate() (#108). A third look-alike source, several trend_start values, is #106.

Found by /code-check on #93; the trend_on case by the plan review for #97.