Tuning at end of King County example does not improve score

[ablaom]

https://alan-turing-institute.github.io/DataScienceTutorials.jl/end-to-end/HouseKingCounty/

cc @vollmersj

[ablaom]

I think we can just add a sentence like,

"Tuning does not improve our test score. It's possible the tuned model does not generalise any better, or that, by chance, we're just doing worse on this particular test set. As the difference is so small, further analysis will likely not settle the issue either way, and we're probably happy to use the simpler, untuned model, or the random forest model with similar performance (or try tuning the random forest model as well). "