[orlp]

Simplifies .sort(["a", "a"]) to just .sort("a").

Also fixes #30081.
Also fixes #30209.

Also includes a simplification pass - we now require the MultipleSortOptions are expanded in the planner.

Generated mostly with AI, verified by me.

[dancsi]

issue (non-blocking): seems like there are some missed optimization opportunities because we run the new rule after sort-collapsing. Here are these cases, as detected by AI

Optimization: Removing a sort the input already satisfies
Example: set_sorted("a").sort(["a", "a"])
Resulting plan: SORT BY [a] is kept, while .sort("a") is removed
OK?: ⚠️ missed
────────────────────────────────────────
Optimization: first/last rewritten to min_by/max_by
Example: sort_by(a, a).first()
Resulting plan: min_by(row_encode(a, a)), while sort_by(a).first() becomes min_by[total](a)
OK?: ⚠️ missed
────────────────────────────────────────
Optimization: Window order_by with a literal key
Example: over(order_by=[lit, c, c])
Resulting plan: row_encode(lit, c)
OK?: ⚠️ already missed on main

[dsprenkels]

nit: I'd make expand_flags() move-based instead of mutably updating the sort_options in place.