fix(tactic/core): remove all_goals option from apply_any

[kim-em]

Fixes the -T50000 regression just noticed by Kevin.

#2245 introduced a regression, essentially breaking set_theory/pgame.lean. This wasn't caught by CI because it's a performance regression, which made solve_by_elim enormously slower on some tasks.

The problem was that I added an all_goals option to apply_any, which told it to try to apply one of its lemmas to any of the goals.

1. This is a useless feature, as apply_any lemmas {all_goals := tt} is equivalent to any_goal {apply_any lemmas}.
2. Worse, I set the default value of all_goals to true. In some situations, this now causes solve_by_elim to try solving unsolvable subgoals in all possible orders, causing a combinatorial slowdown.

This PR simply removes the all_goals option from apply_any. (See point 1!) Note that solve_by_elim*, which tries to solve all goals at once, still works exactly as before, and as expected --- as long as you expect that its behaviour is to first solve the first goal, then try to solve the second goal, backtracking to a different solution of the first goal if that fails, and so on.

The file src/set_theory/pgame.lean now compiles again all the way down to -T6000.

(I also took the opportunity to remove some unnecessary imports from this file, and I promise this is orthogonal to the slowdown issue!)