Swift/TS: Use extracted schemas rather than raw facet Shapes

[fasterthanlime]

Currently all codegen targets (Swift, TypeScript, wire) independently walk raw facet Shapes to generate encode/decode logic. This means each target independently re-derives decisions like is_bytes, named vs anonymous types, transparent aliases, etc.

extract_schemas already makes these decisions correctly and produces a clean, language-agnostic Schema representation. That's the single source of truth — it's already what gets serialized to CBOR and sent over the wire for schema compatibility checking.

The codegen should consume Schema nodes instead of Shape nodes. The flow should be:

facet Shape → extract_schemas → Schema → all targets

Benefits:

• is_bytes, variant classification, struct/enum recognition etc. are decided once, correctly, in one place — not duplicated and potentially inconsistently across every target
• Codegen and runtime schema compatibility checking are provably describing the same type system and cannot drift apart
• Codegen logic becomes simpler — Schema is a clean tagged union with no ShapeKind::Pointer, ShapeKind::TupleStruct, transparent aliases, or other reflection artifacts to handle
• Adding a new language target just means consuming Schema, not understanding facet internals