Firstly, you are correct that on most days, demand peaks around 7pm when solar output is tiny.
Source: http://www.caiso.com/TodaysOutlook/Pages/index.html
However, on most years, demand often peaks around 3pm-5pm, not 7pm (and note that in late summer in California days are longer and solar noon is roughly 1pm, so output is not bad at all). The reason for this is that the days with the most electricity consumption tend to be hot with large AC loads, and those loads tend to be higher during midday.
Source: https://www.caiso.com/documents/californiaisopeakloadhistory...
So when you say solar output is 0 at "peak demand", it's worth mentioning whether you mean the daily peak or the annual peak or a peak measured over another timeframe.
For utility capacity planning, we use the annual peak, not the daily peak, because we want to avoid blackouts most years, not most days. Therefore, adding solar - at today's margin - still does reduce our capex and need for gas peakers.
The math gets a bit more multidimensional when you do this for real and consider aspects beyond peak demand (like ramping, fuel, current generation fleet, tradeoffs in characteristics of new gas plants, geographic spread, congestion, etc.), but the main point remains true and important: the daily peak is not the annual peak, and it's a clear mistake to use the daily peak to conclude that solar doesn't help with peak demand.
In fact, at today's margins, solar both helps with peak power and with energy generation. It's fantastic. (And of course it gets less and less fantastic the more you add.)
Anyway, this is a common misconception I've seen a lot, so I've added a new section to the Duck Curve Wikipedia article you cited above to help explain it.