Why are forecast values slightly lower with `fable::TSLM()` than `stats::lm()`?

Question

I am doing some work which involves modelling values over time, and in the interests of clarity I would like to use the fable package to do this. I want to crea