I like @phonon's answer; but still I guess some sub-questions are not answered.
- Do you just grind them or save them for later?
Depends on the difference between your chess level and chess level of the audience the puzzle is intended to. If the puzzles are from books (esp. course-like structured series), authors will typically mention the level of the puzzles. For example, in winning chess series, Seirawan recommends time to be spent on puzzle and so on. The same is true for other series such as Yusupov's. If the puzzle is not a study, you should try to grinding and if you can't solve it in 'reasonable time', save it for later. If it is a study, one may be grinding against stone; but still that can be instructive (or not). What is 'reasonable time' also varies.
To make the learning process more smooth and enjoyable, I would recommend doing puzzles that are just above your level of play.
2.Do you ever annotate your chess tactics?
Annotation doesn't matter much. But, analyzing all important variations can not be overemphasized. Writing down seems to help it.
IMO, you should set the problem on board and solve rather than solve looking at a figure. That would be a better practice. (I have seen master players pointing out this)
3.Do you try to remember certain patterns and main ideas? Do you categorize your tactics by forks, pins, zugzwang, etc.?
In fact, no need to try to remember things. It is more like getting familiar with it. Practicing with particular patterns or particular themes helps for sure.
IMO, you should mix it up. If you know for sure that you are to use a particular patter only, you can easily figure it out. People (incl. professionals) say that it imprints the pattern in your mind. May be, it does. But it kills all the fun. I would suggest noting down the puzzles and note down some positions from random games (need not be master games) and mix it up and try to solve the position as if you are facing the position in a game over the board.
4.How do you recommend we train tactics puzzles?
This question is answered very carefully by @photon. I don't have much to add; I am only expressing an opinion.
Also go through the recommendations of Seirawan and Yusupov they give on how they expect their reader to try out the problems in their book. You can not go wrong with it. (One such recommendation is fight not only against the problem, but also time as pointed out by @Brian Towers)
Find a set of puzzles that fits your level best (not too easy not too hard; may be a small percentage of them challenging). Just to reiterate, solving over the board is a better practice whether the source is a book or not (eg:a software or website).