First of all, there are a few main categories of exercises I encounter when I read:
1. exercises left to the reader
2. exercises with solutions
When I read a book, I typically dislike exercises left to the reader, especially because when I get to work on a solution, I have no idea whether it's any good at all, especially in a language-specific book where there's a good chance my solution is bad for not being familiar with the language -- not necessarily a lack of reasoning as much as familiarity.
SICP is one exception to this in my experience (I haven't read all the books in your list), because it's not language specific, and I'm reading it with the known mindset it will take months to go through. I'm not reading it to learn Scheme, but to learn to program, and in my opinion, there's a different mindset involved. Still, there were times where I wished I could get solutions without buying a separate manual. I ended up looking for solutions online and found many of them; this isn't an option with many new books, and shouldn't be expected as a feature.
For exercises with solutions, I like these a lot better to begin with. I can work on things until I fail, absorb the solution by reading it, and gaining experience. If you have plenty of exercises, the solutions will significantly increase the volume of the book, though. An online repository is a very decent solution, and even lets you add comments and explanations.
I want to focus on the idea of explanations. Exercises with flat solutions without any explanation of what is going on, just "it works" are something that somewhat annoys me. "Here's code, figure it out kiddo" is a valid approach and makes sure the reader that goes through it still understands a lot, but it doesn't mean the reader won't miss stuff.
If I take the typical quicksort example of functional languages and just say "Implement quicksort using language X" and show one solution, it's up to the reader to figure out whether it's optimal, demonstrative, etc. ("Why is it important to pick a pivot the way it was done?") If there are indications, then it's nice to know how the code works, but you don't always get to understand how to create such code on your own. You will tend to know why the code is the way it is, but not necessarily how to build similar though different code on your own -- there's a subtle distinction there.
This is where a lot of exercises fall short, in my opinion. It's more obvious in math books where only the final answer is given. You can validate your answer, but if you don't know how to come up with one, you're lost. Good luck, ask someone or go back until you're brilliant enough to do this exercise. If there are steps done, nothing tells the reader why that one was picked or why the author decided to attack from that angle instead of any other one. Personally, this would tend to get me into the area where I know how to solve the exercises in the book, but I'm quite unsure about how to solve those outside of it.
I do prefer the approach where the exercises are something you can be guided through. The author explains the design decisions and why things are the way they are. I found it's a format that made more sense to me when I was watching screencasts, or when working with a trainer (or as a trainer), teacher, or whatever. It's how I was taught sports, music, and so on. It might just be a result of how I personally learn better (a series of explanations why things are done the way they are, not just letting me speculate why), and different people tend to learn differently.
The downside of this approach is that it's significantly longer, and you'll have far fewer exercises available as a reader (time and space are limited resources).
That's the approach I tried to take with LYSE. I'm not saying I got it right, or that it's better than whatever other approaches that exist, but it's the one that made the most sense to me as an author, because it made sense to me as a learner. I like that I can be guided through examples, that the person teaching me tells me "here are the possible ways to do it. I picked this one because of reasons x, y, and z", as I feel I now know how to make similar decision on my own in the future when encountering similar problems.
One of the approaches my statistics books take is to include the numerical answers in the back of the book but have a separate guide for sale if you want to know why specific approaches were taken. By being an extra book, students who need the extra help pay extra for the time to compile all the extra solutions. This also keeps the cost low for people who only need a reference.
When you say 'my', I can't tell if you mean books you've written, or books you've read. Either way, writing the solutions is non-recurring work; the cost to print the extra pages should be pretty marginal compared to the presumably > $100 cost of the textbook. To my mind it looks more like 'we get two bites at the apple', rather than saving the poor students the extra $10 to get the solutions with their textbook. Of course, I don't know how much the solutions manual costs, but I feel like this approach compromises the student's experience, when they should really just get all the content up-front.