Sorry for effortposting where it's potentially unnecessary.
Many traditional textbook problems show up on LeetCode: for example, #23, #72, #141, #207, #322. They have textbook answers (of the kind that you're unlikely to reinvent starting from nothing). I could teach a somewhat disjointed introductory course on algorithms entirely from LeetCode problems. I could do that because I can tell which problems are typical, and which ones are distracting, exotic, time-consuming variations. For example, LeetCode #1 is a common textbook technique, #2 is a simple exercise in being careful, #3 is somewhat tricky, and I haven't seen #4 in the wild anywhere outside LeetCode. The solution that LeetCode authors propose for it is complicated and poorly explained.
People do compile curated sets of LeetCode problems. Without curation, you will waste a lot of effort on time-consuming, but less relevant problems. And it's also not clear how you would assess the quality of someone's curation. It's not like university professors release their own lists of LeetCode problems you should tackle. (One possible exception to this is the newest edition of Skiena's Algorithm Design Manual, which makes reference to relevant LeetCode problems here and there.)
The final problem with LeetCode is that the user-led discussion boards are occasionally helpful, but at most times are just awful rabbit holes of people competing on trivial runtime differences and code golfing their solutions. I think immersing yourself in LeetCode comments is a very reliable way to learn dysfunctional coding habits and pick up a distorted worldview of how to approach the coding part of an interview.
