…s unused

The `parse_embedded_nodes` function was silently skipping `SvelteHtmlBlock`
nodes (`{@html expr}`), so the JS expression inside them was never added to
the embedded snippet list. As a result, any variable referenced only via
`{@html variable}` was incorrectly reported as unused by `noUnusedVariables`.

Now `SvelteHtmlBlock` is handled the same way as other expression-bearing
blocks (`SvelteIfBlock`, `SvelteKeyBlock`, etc.): the expression is parsed as
a JS snippet and visited so its identifier references are tracked.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

In Svelte files, <pre> was being parsed the same way as <script>/<style>:
its entire content was consumed as a single raw HTML_LITERAL token inside
HtmlEmbeddedContent.  That prevented the expression-tracking passes in
parse_embedded_nodes from seeing any Svelte blocks inside <pre>, so
variables referenced only via <pre>{@html expr}</pre> or <pre>{expr}</pre>
were still incorrectly reported as unused by noUnusedVariables.

The formatter already has an independent HTML_VERBATIM_TAGS list that
includes "pre", which causes it to emit <pre> content verbatim regardless
of how the parser represents the children.  Skipping the embedded-language
path in the parser for Svelte is therefore safe: formatting is unchanged,
and {@html} / interpolation nodes inside <pre> are now visible as proper
AST descendants.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>