latency regression in optimizer

[JeffBezanson]

Caused by #37884.
Before:

julia> @time using Plots
 84.830424 seconds (7.45 M allocations: 521.182 MiB, 0.16% gc time, 1.48% compilation time)

julia> @time display(plot(rand(10)))
  7.416645 seconds (11.68 M allocations: 676.491 MiB, 2.72% gc time, 95.68% compilation time)

After:

julia> @time using Plots
 85.947123 seconds (7.58 M allocations: 528.318 MiB, 0.16% gc time, 1.57% compilation time)

julia> @time display(plot(rand(10)))
  7.657525 seconds (11.81 M allocations: 688.000 MiB, 2.78% gc time, 95.82% compilation time)

[vchuravy]

Happy to revert. I need it for follow up work, but I can reland it after the branch.

[oscardssmith]

What's the status of this?

[vtjnash]

I think we should just close. We need more significant changes to improve/fix this than this difference represents. But good to have this observation of timing. This is probably something we should track in BaseBenchmarks instead.