inventory the remaining render-equal spelling axes and decide a normal form for each

Ruled by Dan (Sep 5), broadening the heading/fence decisions: every axis where two source spellings render identically gets a DECIDED normal form (asciidoc-native where one exists), not a preserved choice and not a configuration option.

The worklist is already machine-enumerated: the confluence gate exception table (tests/conformance, the render-equal spelling pairs with their mechanism rows) lists every known axis where formatting currently maps two render-equal spellings to two different outputs. Decided so far: bare URLs (landed), block anchor form (in flight), admonition form (in flight), double-quoted positional unquoting (in flight), tilde open blocks (in flight), markdown headings and markdown fences (issues filed alongside this one). Candidates the table or the pre-spec suggest and nobody has ruled on: setext vs = headings, thematic-break spellings, table delimiter families, list marker respellings, attribute-entry spacing, xref macro vs shorthand.

Deliverable: one short section per remaining axis - the two spellings, the render-equality evidence, the recommended normal form, and the conversion-domain hazards - filed as one issue per axis after Dan confirms each recommendation (the survey itself decides nothing unilaterally). Each conversion that lands retires its exception-table rows; the definite-end condition for this issue is an empty unruled-axis list.

Start with the confluence gate exception table in tests/conformance and compare its render-equal spelling pairs and mechanism rows with the pre-spec. For each remaining axis, document both spellings, the evidence, a proposed normal form, and conversion-domain hazards, then get Dan's confirmation and file one issue per axis. Done means the unruled-axis list is empty.