You don’t need to do any tricky cursor stuff, because it’s all semantically wired up already.

I’d say if the form is short and an obvious pattern (like sign up or log in), you could use the placeholder visual pattern, but use real labels instead.

If they click the area taken up by the label, it will activate the input. If they click the input, it will activate the input. Both correct.

I like to keep the ‘verb’ + ‘noun’ structure when writing button labels — this makes the action more prescriptive, e.g. ‘Save post’, ‘Next step’, etc, as opposed to ‘Save’, ‘Next’