refactor(linear_algebra/affine_space/finite_dimensional): change three lemmas to instances (#16331)
I realised after the PR with these lemmas was approved that they are
actually suitable to be instances (don't depend on any hypotheses that
typeclass inference can't deduce), and making them such sometimes
allows removing explicit `haveI` uses because typeclass inference can
then find them automatically.