A few basic questions

I like zeroing out thrust every turn. My feeling is this is both the simplest model and the one that any moderately advanced bot will use anyway (most Robocode bots recalculate movement every tick), so why not keep your API simple and make it the only model.

My first thought is that drag sounds not fun and not that interesting a problem either, but there are good reasons you might want it - e.g., to keep max speeds in check, if you think that would be good for gameplay.

Your thoughts on missile behavior sound interesting and like a big change from Robocode, which is cool. Will there be any limit to what you can see on radar - e.g., walls obstructing it, distance, etc? That would play interestingly into the invisible mine thing, like being able to move your stuff around when you think the enemy can't see it. Incomplete information is a pretty powerful gameplay dynamic, and without it you risk creating a "solvable" game, especially with computer players.

I think my biggest advice is to be vigilant about only adding complexity that really multiplies the gameplay depth. I posted a bunch of thoughts about this in BerryBots vs Robocode game rules.

Chess is a good example of a game that probably gets it just right. Each type of piece multiplies the complexity, and the result is a game where the greatest human minds, with tons of practice, can almost play it perfectly. Any more pieces and it would become too complex and even less approachable, and any fewer pieces and it probably would have become uninteresting ages ago.