Yes, of course it’s about the amount of delegation, not number of delegators. I don’t think there is any implementation in Cosmos that calculates active set inclusion based on the number of delegators, but happy to be proven wrong 
Simple cap on the max tokens is a good idea, but I guess popular validators can just spawn new nodes, and gobble more of the 175 active slots. The re-delegating idea definitely needs to be worked out more, but I think it may be interesting to model.
As for inactive validators, I don’t mean inactive validators without any delegations, but if you look at the 175 set, you see there are sometimes shifts between no. 175 and no. 176, especially when there are some downtime and the active validator gets jailed. If a validator does NOT want to be in the active set, I think they can just turn off their machine, then they won’t be added into the active set. No one will force them into the active set then.
About “financially stable”: There are fixed costs to running the node setup depending on each validator. Cosmos is a popular network and I sincerely doubt there will be a lack of validators who wouldn’t take the place of validators that will drop out. Also… personally I don’t like to identify potential bankruptcies without clear data. If a validator is going to face bankruptcy now because they can’t meet cash flow, that means to me they are already in significant liabilities (possible if they bought 400k hardware in their DC), is this the case? Or that they would just not be able to cover their costs, of which they could stop validating and eating into their assets (unless they have fixed contracts on DC or cloud that they will still have to pay regardless).