but shouldn't I be the one to decide if I include a @reply or move a discussion to chat?
This sort of thing is not a black-and-white issue. There's a range between extremes, and SO and its siblings are seeking a particular balance that Jeff & co. believe is optimal.
If you think leaving things almost entirely up to the users is preferable, there's always Usenet or 4chan.
I fear that decisions are starting to be made based on what's best for the system, not what's best for the users.
What's best for the site is by definition what's best for the users. What an individual user thinks is best for themself, however, is not necessarily best for anybody, including (from a broader perspective) the user in question.
If you find the above assertion strange I would encourage you to look into behavioral economics and game theory; these principles are well-known.
Ok, I'll spell things out: What people want to do in the short term at small scales may be "better" for them in some sense, but if the result of this (in aggregate) reduces the quality of the site, this harms everyone, including the users who simply chose what they thought was best for themselves.
Reducing the amount of noise on the site is a major part of this, from discouraging commenting to locking popular "fluff" questions to aggressively purging low-quality posts. Everyone benefits from keeping the overal quality of SO as high as possible, and this requires frequently preventing users from using the site in ways they might prefer.
Arguing that the decision to do or not do something should be left to the individual user's discretion is a nice idea, but it doesn't work in the presence of actions that benefit individuals at the expense of minor harm to the entire population; without some sort of constraint the result is inevitably a rush to the bottom as everyone does the "best" thing locally and collectively makes things much worse globally.
Making the system subtly coercive is one way to deal with the problem; rather than restricting outright, make the right thing the path of least resistance. Centralized authority is another way, but it scales poorly. Community norms are the best solution, but requires convincing enough people cooperating that the individual benefits of globally suboptimal choices are outweighed by the attached social disapproval.
All of the above are well-established concepts, both derived formally from the mathematics of game theory, and from empirical observations in behavioral economics. Worrying about things "moving from helpful to coercive in how people use the system" or fearing that "decisions are starting to be made based on what's best for the system, not what's best for the users" are often a sign of a naive way of thinking that ignores the existence of non-positive-sum interactions.
In computer science terms, letting users do as they please is a greedy algorithm for optimizing global success. Most social contexts are an optimization problem not well-suited to this approach.