Metamorphic TOML testing: one tree, many equivalent renderings

To test TOML table semantics without a majority vote, generate a random tree (tables, arrays of tables, scalars) and render it in randomly chosen equivalent ways: [header] sections, dotted keys, inline tables, [[aot]] vs inline arrays, implicit super-tables, sections out of order. Every parser must return exactly that tree. Use a random topological sort for section order, with two constraints: [[x]] elements stay in order with their sub-sections between them, and a header under a dotted-key table must come after the section that defines it (otherwise you hit the spec-ambiguous 'dotted key extends a header-implicit table' case). Gate TOML 1.1 syntax (newlines and trailing commas in inline tables) per document and skip 1.0-only parsers on those. Byte-level ddmin wanders off on these docs (it shrinks keys to empty "" and lands on unrelated empty-key bugs), so reject candidates with empty keys or reduce by lines and elements instead.