web: no engine reads the HTML and CSS relations the parser emits #1908

The parser already reads .html/.htm/.xhtml and .css files into seventeen web relations (all-html-*.csv, all-css-*.csv, schema under parser/src/schema/web/), but no engine reads them. When you index a plain HTML + CSS + JavaScript site, its pages, elements, classes, ids, stylesheets, rules, selectors and custom properties don't reach any graph, so context, impact and path can't answer about them. A developer porting such a site to a component framework has to grep for every class and id. Scope: HTML and CSS only; JavaScript files and inline script bodies are not parsed by this layer.

Reproduction

<!-- index.html -->
<link rel="stylesheet" href="widgets.css">
<div class="widget-card" id="main-widget"></div>

/* widgets.css */
.widget-card { color: var(--widget-fg); }

The parser writes html_element, html_class_reference, css_selector_part and css_declaration rows for these files. No engine declares those relations, so no graph table receives them.

Root cause

• parser/src/extract.ts copies the web tables into the java/, javascript/ and typescript/ IR folders, but no engine reads them. There is no graph/web/, no web adapter in graph/bundle/languages.ts, and no web member in the Language union.

$ grep -n "^export type Language" graph/bundle/schema.ts
18:export type Language = 'java' | 'typescript' | 'python' | 'javascript' | 'csharp';
$ grep -rln "html_element\|all-html-" graph/
$

Checklist

• graph/web/ engine: HTML and CSS as one web language, --lang web, with in-web edges. A selector that can't be decided statically is recorded as unknown.
• Bundle web adapter, documented in graph/bundle/SCHEMA.md. context, impact and path answer on web nodes.
• Nodes shaped for conversion: shared structures across pages, resolved styles per element, tokens, breakpoints, forms, assets, unused CSS.
• Torture suite under graph/test/web/ against an independent oracle, and measurement on real multi-page sites.