This might not be as much of a problem as you think. Why? Because munchkining, minmaxing, optimising, whatever you want to call it - is severely limited in 5e. The main techniques for it in previous editions of D&D involved things which are significantly less effective in 5e.
Multiclassing has been crippled by the all-important ability score increases/feats being a feature of class advancement instead of character advancement. There is currently a limited selection of classes and feats, so taking advantage of obscure classes, prestige classes, variants, and feats is no longer an option.
D&D 5e also introduces the concept of 'bounded accuracy'; see here for a good explanation of this idea. There is only so much it's possible to do to optimise a character in 5e, and the gap between an optimised character and an unoptimised one will be fairly small.
Your players will still spend time optimising their characters, of course, but that's not necessarily a bad thing. If a player hasn't spent any time on their character, that's a sign that they may not be particularly invested in them.
As far as creating playable characters with reasonable backstories goes, this is a great opportunity to use their munchkin-ness against them. They're going to want to choose a certain background for the proficiencies it offers - make them justify it. You want to be a Sailor who became an adventuring Wizard with a single level in Cleric? That's fine, but you'd better have a damn good explanation for it.
And using their munchkin-ness against them is a great technique to make them roleplay, too. When they create their characters, they'll choose a bond, a flaw, an ideal, and a trait. Let them know that you're happy to hand out inspiration (which is incredibly powerful - advantage to a roll of your choice? Sweet damn, who wouldn't want that?) if they play their character. Positive incentives have been used to motivate people to do what the motivator wants for years, because it works.
Quick bit of backup for all this - I'm DM-ing 5e for a group consisting of 3 munchkins and 2 roleplayers. One of the munchkins is so bad he walked into a core-only 3.5 game and insisted on creating an Artificer. And you know what? He's playing a single-classed Fighter, roleplaying as much or more than the roleplayers, and (as far as I can tell) having a great time doing it. (Oh, and he also has by far the most extensive backstory, he really got into it when he was choosing his background options.)
I'm going to assume the silliness and the need for the plot to be this tightly on the rails are necessary, and that this is specifically a question about how to communicate that this particular technique is making you feel less, not more, invested in the game and the world.
So I suggest this angle:
If you must restrict our agency, don't do it by getting between us and our characters. Do it by getting between our characters and the world.
If he doesn't see the difference between the two, try examples. Imagine a reason you're going into a dungeon:
- A ritual is being held inside, and you must stop it or else the world will end.
- An overwhelmingly powerful enemy chased you or a natural disaster herded you in, and this is your only shelter from them. Alternately, a magical effect is controlling your body in order to compel you to enter.
- Same magical effect, but instead of compelling the character to enter (whether you like it or not), it compels the character to like it.
Mechanically they're extremely similar, but thematically you can immediately spot the difference:
- The first has both player and character wanting to go there (provided you started on the same page).
- The second forces the character to go there but at least it's understood that the world is imposing on the character. They retains the agency to at least feel how they want about it, or to attempt to resist it (even if resistance is impossible).
- The third takes a degree of control away from, not the character, but the player. At that point, why is the player even at the table if the DM is going to exercise that level of control?
Now, the point at which magical compulsion crosses from #2 to #3 is going to vary with different players and groups, but the most common hard limit I've seen is affecting what a character thinks about what is going on. But again, this is something everyone needs to be on the same page with, and hence, needs a degree of player buy-in to make it work. It sounds like this wasn't the case here.
Best Answer
Well, you slightly created a rod for your own back here, but strange things happen in D&D!
There is no specific rule for turning a flesh golem back into separate people (funnily enough). But the characters are dead and their body parts make up the golem. Killing the golem would allow someone to separate out the body parts (I recommend a good axe).
Then a cleric could cast Resurrection on the appropriate parts. This spell doesn't seem to say how much of a body is required, but it does say it closes all mortal wounds and restores lost body parts. So I think allowing it to work on half a body would be acceptable (failing that, True Resurrection doesn't even require a body, so that would certainly work).
(Resurrection would be too high level for the party to access cast themselves, but a DM that really wants to provide the characters with an 'out' could always allow an NPC cleric to be sought out. Of course, such services don't come cheap, and the cleric may want a service in return...)