Normalizes Elixir AST for structural comparison.

Transforms an AST so that structurally equivalent code produces identical output regardless of variable names, metadata, or (optionally) literal values.

All normalizations are performed in at most two AST traversals:

Pass 1 (always, single fused walk):

1. Metadata stripping — removes line numbers, columns, counters, and other compiler metadata from every node.
2. Boolean operator canonicalization — &&/||/! are rewritten to and/or/not so stylistic choice between short-circuit and keyword operators doesn't affect comparison.
3. Sigil expansion — ~w(foo bar)a is expanded to [:foo, :bar] (and likewise for string modifiers) so sigil word-lists match their literal equivalents.
4. Pipe normalization (optional) — x |> f() is rewritten to f(x).
5. Variable normalization — replaces variable names with positional placeholders (:$0, :$1, …) based on first-occurrence order.

Pass 2 (abstract mode only, single recursive walk):

6. Literal abstraction — replaces concrete literals with type-tagged placeholders to detect Type-II clones.
7. Map/struct field sorting — sorts key-value pairs so that %{b: 1, a: 2} and %{a: 2, b: 1} produce the same hash.
8. Guard abstraction — in when clauses, replaces all function/macro call names with a :__guard__ placeholder so that when is_binary(x) and when is_atom(x) produce the same hash. Covers all Kernel guards, Erlang BIF guards, defguard macros, and library guards like Integer.is_even/1.