Here's the two ways I came up with in the end:
Way 1 uses a through hole sized to be wide enough to take the entry end of the connector, but narrow enough to catch on the wider back section of the connector. The connector is slid in 1A/1B, with solder applied to one side, allowed to cool, then to the other side, and a blob of bluetack to hold it steady before the first side's joint is done. Pretty messy overall, some solder leaks down the space between the connector's brass body and the sides of the through hole, and it seems that when one uses as much solder as is needed for things this big then truly cleaning off all the flux residue afterward is impossible.
Way 2 uses a >7mm x 7mm pad with two holes in it. Two stripped singe core wires are looped through the hole in the back section of the 3.5mm female banana/bullet connector and pushed down through the two holes in the PCB pad (2A/2B). The wires are twisted together below to tighten the connector down, 2C/2D. Then apply solder above (2F) by heating both the pad and the connector. Then cut off the wire ends and solder below (2E) heating both the cut off wire ends and the underside pad. Again, couldn't clean all the flux of (rosin, normal sort not activated or super-activated, flux being from the core of leaded 60/40 solder reel, no extra flux added) however hard I scrubbed with a brush and IPA (99%) or water (tried both alternately).
The top left image shows the board with 3.5mm banana wires plugged in to those method 2 connectors.
I think method 2 is more promising, and could be strengthened by adding some extra holes in the pad for further wire loops run in other directions (like over the top of the main barrel of the connector) to be run and twisted tight before soldering.
EDIT: forgot the image attachement, here it is now