Nuclear, you can spin down a plant and balance generating capacity elsewhere.
Bridges, you can provide access via detours or ferries.
Almost every other mega-engineering project we have is network related: in that there exists alternate connectivity.
Dams are intrinsically linked to their topography and hydrology. Where can you source replacement water at that volume? Where can you put excess water? They essentially have to be hotfixed in-place, while in operation.
Drain down, dig a bypass, temporary dams inside both sides of the bypass, do the offline / bypassed work, reverse the procedure. Maybe the second time we'll have the foresight to make the bypass more easily serviceable again.
Btw, the time scales are quite long even if it can be done. Filling Lake Mead took from late 1934 to may 1937 or about 2.5 years. Assuming the downstream infrastructure can take double the normal flows, you'd need 2.5 years to drain and another 2.5 years to refill after the maintenance is done. The opportunity costs are huge, it seems at least worthwhile to check if you can do the works underwater with robots or something like that.
"we always forget that maintenance in perpetuity is much much harder"
I imagine that part of the problem is that the heavy maintenance for large engineering projects begins years or decades after completion. Upon completion, it's hard to imagine these huge creations falling apart; add to this an element of "this won't be my problem (I won't be here in 50 years)" and you've got a recipe for short-term thinking - especially by those who don't have the expertise (read: the politicians and bureaucrats).
[1] https://www.marketplace.org/2015/05/19/americas-infrastructu... [2] John Oliver: https://www.youtube.com/watch?v=Wpzvaqypav8