The default in hypothesis is to use the normal distribution (df = Inf) so the modification below results in identical p-values.

(mepvalue <- marginaleffects::hypotheses(mod, "b4 >= 0", df=df)$p.value)
all.equal(pvalue, mepvalue[1]) # TRUE