Pure model layer for x-code — a code-display component with an in-house,
regex-based syntax tokenizer.

No DOM, no side effects. Every function here is a plain data transform and
is exercised directly by model_test.cljs with sparse inputs.

The tokenizer is deliberately lightweight. It covers js / json / css / html
only and degrades any other language to plain text. It is a highlighter,
not a parser: it does not track nested languages, and a handful of
context-sensitive cases are knowingly imperfect (see docs/x-code.md).

DOM + lifecycle layer for x-code — a code-display component.

The component is stateless: DOM = f(attributes, properties). Source code is
read from the `code` attribute or else the element's light-DOM textContent,
tokenized by the pure model layer, and rendered escaped into
the shadow `<code>`. A MutationObserver re-renders when the light-DOM text
changes. The expand/collapse state lives in the observed `expanded`
attribute, so it too is a value, never a mutable field.

Instance fields hold only the refs map, the cached model and the
MutationObserver.