The definition of Rambo should be the situation where it's more tactical sound to use 1 unit, to own all other classes. And only with a cav unit is that is possible.
You can rambo with any class and it is not limited to cavalry. Cavalry is simply the most effective class to rambo with, not the only class. You should see some of the players who use solo shock troops to rack up 6-7 kills by running off alone by themselves before dying.
Any time that you separate from your unit to either:
1. Kill enemy Ai without the presence of your troops.
2. Distract enemy Ai without the presence of your troops.
-Turn Ai's shields around without the presence of your troops.
-Soak up archer fire without the presence of your troops.
-Lead Ai who are charging to follow you away from the main fight with a single unit.
This is what Rambo is, and the exploits that result from it.
As Callum stated, you are never going to be free of these exploits by simply improving the Ai because the Ai can only ever be so good. No matter how good the Ai gets, the player will always be smarter and able to abuse it's targeting mechanics.
Take for example: "Turning an Ai's shields around with a single unit"
If you program the Ai so that it ignores the single unit threat that is within proximity, then the player is free to get behind the ai and kill them.
If you don't program the Ai so that it ignores the single unit threat, then the player will be able to turn the shields due to their proclivity for "Self-preservation", effectively distracting them and rendering them unable to function properly.
In either scenario the enemy formation is at a loss, and all that it cost the player is potentially 1 life. The point is that Ai will always have a behavioral pattern that will be easily identified and exploited by the player no matter how good you program it to be. Leashing doesn't prevent you from turning Ai shields, it prevents you from doing it with little-to-no risk by putting your friendly troops in harms way which is a more fair exchange.
I would also be in favor of restricting Captain respawns to 1-3 lives as it achieves the same thing, you will have to weigh the potential cost of the engagement (potentially dying and not being able to respawn) with the benefits (distracting or killing enemy ai) and then make the decision of whether or not you deem it worth it in any given scenario.
The problem with rambo tactics is that there is no real risk in using them, and that is why the most "efficient" use of your troops is to use them as extra lives. (If you can get 6 kills for every 1 life that you expend that is a good trade for you) There needs to be some form of risk added to using these tactics, which is why Leashing or Restricting Captain Respawns are such popular suggestions because they facilitate this risk. Leashing makes it so you risk your ai troops when you engage in combat. Restricting Captain Respawns risks you being shut out of the fight if you die while fighting. Both will incentivize the player to avoid ramboing and thus will push the gameplay in the proper direction of controlling and commanding Ai troops to work as an army to defeat the enemy army, which is what captain mode should be.