OK, the bar option works pretty easily, actually, with some handle-diving for how things are organized.
To use bar to get scaled as between min/max for the second bar, the bar definitions are intervals [0 L] and [L H-L] so the sum or max of second bar in the stacked plot is L+(H-L) = H instead of H+L if plot L,H as normal.
hB=bar([y(1,:); diff(y)].', ...
ends up with
In the end, this doesn't seem much more effort to fix up actual temperature min/max data by the difference than would be to compute the error values for the errorbar solution. It's a little more abstract in needing to set the color for the one bar so with Steven's observation of the new property for the error bar ends lengths being able to be set, that's probably the better solution.
BTW, the idea of simply
to clear the first bar instead of rewriting the 'CData' doesn't work-- bar then turns the plot into a standard bar plot with zero baseline; doesn't keep the 'stacked' property any longer.