Spinning that thread a little further:  If such a city is attacked, a militia would be formed depending on how many figures from disbanding military units went to the barracks.
A second integer to save per city. We're getting excessive now!

It won't be some crack elite force but enough to drive off the odd wolf or bear.
This reduces the micromanagement of garrisoning but by itself won't stop or even delay an attackig army.
And you'd need barracks for that to work so it doesn't give you auto-defense when spamming lvl 1 cities all over the place.
And to get any such militia, you must have trained troops before and disbanded them, losing their utility as fighting troops.
If someone were to get flashy with such a system, the equipment quality of disbanded troops would somewhat reflect on the militia units.

I'm not entirely happy with this idea, yet.  Something isn't right but I can't put a finger on it.

Here's how I see it:

There needs to be a unit manager UI (remember the fleet manager?) to manage existing troops.  In this UI, there would be formation, and individual units.  If you want to combine units, you have to re-train/equip them with all the same stuff, then you combine them into a unit (1 turn or something to train as a unit).

The way I see it, the XP of the troops is the thing that needs to be preserved the most.

It's perfectly believeable that the troops need to re-train with the new weapons and equipment so the time to train the new unit in the city could just be left as is.  The benefit would be not having to go through  the many many fights that you fought to get the unit to a high level.

Yeah, this retraining makes sense. But the XP reallocation doesn't. If we're representing that the same dudes go back into the force, each individual person's XP should be stored, and people with the most should be recruited first. Then the XP for large groups should be averaged, with no unit having less XP than the level you selected before training. This has the flaws that it's a very fiddly system on the users end, and that it would probably be a hassle on the programming end as well.

However, we could also do it somewhat like Lords of Magic. In that game, you could position experienced champions in any unit-training facility, and they would increase the experience with which a new unit would be made. In Elemental, we could represent that by having each unit decommissioned at a given city be used for training, either giving all new units an exp bonus equal to maybe 10% that of the most experienced decommissioned unit, or boosting the trainable XP (unlocking higher levels of training if necessary) by perhaps 50% of the experience earned by the most experienced unit. Additional units beyond the most experienced could further increase that by 10% of their own experience, or could decrease the straining time added by heightened level. Of note, this would encourage city specialization, since very skilled units are of course at a premium, and would be required to maximize the skill of new recruits from a given city. This may not be perfect, though I've not noticed any inherent flaws, and isn't totally in keeping with Brad's philosophy of all the resources being global. Nonetheless, it seems to me that it's the optimal solution here.

I agree that it's not the optimal solution, though.

Or add a couple of options

disband unit - XP lost, half the gold and materials used in the equipment are regained

Retrain unit - same calculation as for training a new unit but only for the equipment elements that are added, any equipment that is removed, half the gold and materials are recovered. XP of original unit is retained.