Samsung's 28Ghz solution was rated at something like 1500ft at maximum power. So one per city block might be more realistic than one per lamp post but it will ultimately come down to capacity. The mmWave portion of 5G is designed to high density deployments.
The problem is that omnidirectional (so non-directional) radiation causes the problem, not fixed point-to-point links. So, with the omnis the best idea is to limit their TX power and spread the noise via ultra-wide band multiplexing (using random orthogonal coding [which is CDMA or lately frequencies (subcarriers/subbands), which is OFDMA] minimizes the interference), but with the p2p links, you can go wild, use a relatively high power high frequency narrow band high symbol/baud modulation (like QAM256 - used in 802.11ac wifi).
But of course cables rule, because they are really great at containing the EM radiation. And lamps already have cables to power the lamp, and the micro-nano-pico-cell thingie will also need power too, and then you can just do Power over Ethernet and be done with it, and then concentrate the copper wires and switch to fiber.
Though probably the p2p mesh backhaul module would be pricey, but sometimes getting your cable to a switch is equally problematic/costly, so that's why probably a bit of both from a cost perspective too.