This is probably better as a comment, but too lengthy.
Abbott is a junior level analysis text. Baby Rudin is definitely a senior level analysis text, designed for a two-semester course. Rudin is terse and makes you work for it. I felt the book was too difficult when I was going through the course; but in hindsight, it was appropriate (which means I grew from the course!).
Senior analysis has the burden of building one's mathematical maturity for graduate studies. When I took senior analysis, we were expected to put in 20 hours each week. There were several grad students from other departments (at least for the first semester), that were taking this course to improve their writing skills for publishing research papers. From this perspective, Rudin was appropriate.
Measure theory is usually taught as a graduate course. It doesn't take much work to define a measure and derive basic properties; only basic set theory and basic algebra are involved ($\sigma$-fields are basically free algebras). Once you start defining the Lebesgue integral or talking about modes of convergence, having a stronger analysis background will be helpful. Graduate level probability theory classes usually introduce the requisite measure theory upfront (from what I have heard). So if your goal is probability, I would try and find a graduate probability theory course website or text to help guide you.
I would personally suggest Rudin for the rudiments of analysis. You will definitely want to cover series and sequences, point-set topology, continuity, differentiation, and series and sequences of functions. It may be safe to skip the derivative as a linear operator and the Riemann-Stieltjes integral (though these are always good things to know, especially thinking about the derivative as a linear operator, if you are doing machine learning and stochastic optimization). I would avoid Baby Rudin for measure, as I have not heard good things about his exposition.
Trench is another alternative to Rudin which I have heard recommended. I am not familiar with this text personally.
If you find Rudin too terse, you can always pick up Abbott.
Bartle's The Elements of Integration and Lebesgue Measure is a friendly text for measure theory, though it does assume some maturity. Again, measure is really a graduate level topic.
Also, if you are still a student, it might be worth sitting in on an analysis class at your institution. Having someone with intuition and classmates with whom to converse will definitely be helpful for learning the material.
I read first few chapters of Folland's Real Analysis during summer after my senior year at Ohio State to prepare for a rigorous PhD program that I was soon to begin (with one focus being pure / applied econometrics). I had previously also read Baby Rudin, ch. 1-7 for a course (you'll need ch. 7 since uniform convergence of sequences of functions is important for measure theory). Folland is a bit terse, but precise, and it has been the primary text for first year grad analysis at many universities because its coverage/presentation is fairly traditional and complete: measure theory, basic functional analysis, topology that you need for analysis if you haven't taken a course in general topology), Fourier analysis, distributions. And one of the other reviewers is correct: you need to read at least Ch. 1 - 6 (to get Lp spaces), and Ch. 8 (Fourier Analysis) is very helpful for those who will take probability theory like you). Ch. 7 is a "bonus" since almost no text at this level proves the major theorems in locally compact Hausdorff spaces as thoroughly as Folland (and you'll see other measure theory texts refer you to this chapter for its coverage). I agree Folland is a bit dry, but adequacy of coverage makes up for it. A more lively presentation of similar topics is first 3 parts of Serge Lang's "Real and Functional Analysis" (and part 4 on differential calculus in Banach spaces is well done too). I just found about Axler's book, and I'm looking forward to skimming it soon.
Best Answer
In my opinion there is no really intuitive argument, you have to do some calculations. But very simple steps are sufficient to show that $J = (x-\epsilon, x+\epsilon)$ contains only finitely many elements of $K$.
Certainly no number $\frac 1 n$ is contained in $J$. So let us consider the numbers $a_{m,n} = \frac 1 m + \frac 1 n$ with $n \ge m$. By definition $\frac 1 m < a_{m,n} \le a_{m,m} = \frac 2 m$.
For $m \le p$ we have $a_{m,n} > \frac 1 m \ge \frac 1 p > x+\epsilon$. Therefore only for $m > p$ it is possible that some $a_{m,n} \in J$.
For $m \ge 2(p+1)$ we have $a_{m,n} \le \frac 2 m \le \frac{1}{ p+1}< x-\epsilon$. Therefore only for $m < 2(p+1)$ it is possible that some $a_{m,n} \in J$.
Let us consider the finitely many $m$ with $p < m < 2(p+1)$. For each such $m$ we have $\frac 1 m \le \frac{1}{p+1} < x-\epsilon$, hence all but finitely many $n$ satisfy $a_{m,n} = \frac 1 m + \frac 1 n < x -\epsilon$. In other words, at most finitely many $a_{m,n}$ can be contained in $J$.
Clearly 1. - 3. imply that $K \cap J$ is finite.