Is it bad practice to merge single player code with multiplayer code by making single player a client connecting to a private, automatically started, local server?

Afaik its standard practice. Coding a single player game from the start and then turning it into multi player is really hard, on the other hand, going from multi player to a single player game (in the way you proposed) would be easier to do.

That tradition goes a long way back. Original Quake did that, too, and I'd be surprised if it was the first. It's a good practice.

Quake, Quake 3, Unreal Engine games, Crytek's, even some source games, these work on the principle of a single player game on a loop-back / virtual local network.

Usually, the client / server separation is very strong. Quake 3 is an example of that. The server part is basically a renderless dedicated server, the client part handling the rendering, audio and inputs. Both completely separate contexts, if you will (code-wise, two seperate projects / dll / libraries). It also means, seperate contexts for each entity types, server-side entities that handles logic, and client side entities that handles audio and graphics with virtually no logic of their own (apart for stuff like client-side prediction and particle physics). Plus the networking, messaging and serialsation gubbins in the middle mediating between the two.