Trend a metric across a spec's whole version history (tree-aware).

Where 'dow compare' contrasts exactly two versions, 'dow trend' follows a metric
across ALL versions in commit order, so a slow drift over many iterations - not
just the last hop - is visible. It reports every version's value together with the
change since the previous version and since the baseline (v1), and labels each hop
as a baseline, a same-config re-run, or a config change (the same tree awareness as
'dow history').

Two kinds of metric are trended: the built-in text stability (when the spec embeds
output text) and your own evaluator scores saved under evaluation.metrics. With no
--metric, every numeric metric found is shown; pass --metric NAME to focus on one.
dow computes no new numbers here - it only lines up the ones already recorded.

With --plot, the spec's evaluation.plots functions receive the trend series (kind
'trend') and dow stores the figures they produce as content-addressed artifacts
(dow ships no plotting library).

@examples
dow trend
dow trend --metric stability
dow trend -m accuracy --plot
dow trend --spec prompt_sensitivity -m krippendorff_alpha
