Document how CIDER picks the form a command acts on

[bbatsov]

Stacked on #4173, so the diff to review here is just the two doc pages.

The terminology section named last-sexp and sexp-at-point but never said what follows from them, so the rules had to be inferred command by command. There's now a section that spells out the three ways to say which form you mean, a table of the at-point variants, and a snippet you can paste to rebind the flagship keys to them if you prefer pointing at a form over standing after it.

It also records why the cursor goes after the form rather than on it, which until now was folklore: that's how Emacs Lisp's own eval-last-sexp works, and SLIME's and SLY's, and following that prior art has been a deliberate choice rather than an accident. The keybindings section already says as much about keys; the same reasoning governs form selection, and after the recent round-trip on this it seemed worth writing down.

Macroexpansion gets its own section, since it's the one operation that resolves its target differently: an expansion needs a call form, so a bare symbol widens to the call around it and an opening delimiter expands the form it opens rather than the parent.

Two things I have not done here. There are no GIFs illustrating the selection modes yet - they want a live REPL to show real evaluation overlays, and that harness is worth building properly rather than faking the illustration; I'd like to do it as its own PR. And I'd like to look at what SLIME, SLY, Geiser and the Emacs Lisp modes do that we don't, since we've only ever borrowed the convention itself and there may be more worth taking.