Consolidate stats handling

[avik-pal]

As pointed out by @oscardssmith, it would be good to store a centralized stats in the cache and update references to it instead of adding up the stats at the end.

Would probably make SciML/OrdinaryDiffEq.jl#2167 easier

[oscardssmith]

Also worth noting that I'm pretty sure that for the polyalgs, the stats are just wrong. They appear to only report the stats from the solver that it actually used: https://github.com/SciML/NonlinearSolve.jl/blob/b389d0e9e9e4a1ff4350a2cce824ec8a03310e53/src/default.jl#L184C9-L184C14.

[avik-pal]

Yes that is an unfortunate decision I made. Previously we showed no stats but then I decided it is atleast worthwhile to get the stats for the oslver that worked. Using a reference to a top level stats object solves all of them