This would require flexible bot shapes, because cells interact over an area. Two circles can interact only over a single point.
But we can have some intermediate solution. The simplest one would be to that a multibot gets still shown as a circle (with an area representing the sum of areas of fused bots), but certain portions of its border are designated to individual bots. The exact rules for movement, shots, energy sharing, damage distribution etc can be worked out. This is a very simple solution and can be introduced fairly quickly, because for such a multibot almost everything (including physics) is going to be the same as for current single-bots.
A step towards more advanced things would be for us to agree on certain pre-determined shapes for bots, based on how many edges they have (triangular - 3, square - 4, and so on up to a certain limit). Then we define a number of ways n which these shapes can hook up to each other and allow bots to choose from these pre-defined scenarios.
I am not too hot about these ideas myself, but they might work...