fix(security): reject every markup-bearing data: media type and legacy script scheme in the MDX scanner

[hivecommons-hive]

Security Fix

scripts/validate-architectures.mjs gates every imported third-party
architecture body with findActiveContent() from
scripts/lib/mdx-active-content.mjs. Its DANGEROUS_URL_PATTERN recognised a
strictly smaller set of schemes than activeScheme() in
scripts/lib/svg-active-content.mjs, which guards imported SVGs against the
identical threat.

Before:

const DANGEROUS_URL_PATTERN = /(?:javascript|vbscript):|data:text\/html/gi;

Two gaps, both in that single constant:

• Only data:text/html. A browser parses image/svg+xml,
  application/xhtml+xml, text/xml and application/xml as documents and
  runs <script> in them, so the identical payload passed under a sibling
  media type. This is the rationale svg-active-content.mjs:683-688 already
  states in its own comment (added by fix(security): reject every markup-bearing data: media type in activeScheme #1168) — only this file had not caught
  up.
• Only javascript: and vbscript:. ACTIVE_SCHEMES
  (svg-active-content.mjs:32) also carries livescript: and mocha:.

Verified against main @ 593b9fe with base64 payloads, so no raw <element>
is present to trip the unrelated element scan and the scheme test is the only
thing that can produce a finding:

FLAGGED data text/html        script-capable URL scheme
MISSED  data image/svg+xml
MISSED  data application/xhtml+xml
MISSED  data text/xml
MISSED  data application/xml
FLAGGED javascript:           script-capable URL scheme
FLAGGED vbscript:             script-capable URL scheme
MISSED  livescript:
MISSED  mocha:

What changed

Exactly two files, no behaviour outside the scheme test:

• scripts/lib/mdx-active-content.mjs — widen DANGEROUS_URL_PATTERN to the
  set svg-active-content.mjs already rejects, with a comment pointing at
  that module so the two stay paired.
• tests/mdx-active-content.test.mjs — two new tests covering each of the
  four media types and both legacy schemes, including behind the character
  references normalizeSchemes() decodes.

Checks

• node --test tests/mdx-active-content.test.mjs — 31/31 pass.
• npx prettier --check on both changed files — clean.
• Widening a gate can only break the build if existing content matches it:
  grep over docs/, blog/, data/ and src/ for all six newly gated
  forms returns nothing, so the validator stays green on the current tree.

The remaining unit-test files fail in this environment only because
node_modules is absent (Cannot find package 'react'); that is unrelated to
this diff.

Closes #1218

---

Filed by sec-check agent (ACMM L4/L5 — hold-gated mode). Hold-gated: human review required.

— hive: agent=sec-check backend=copilot model=claude-opus-5 copilot=1.0.88

[hivecommons-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "sec-check" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will keep the hold label until a human removes it. Operators can make a deliberate one-off release during an ACMM level change with release_level_holds=true, but level changes never release this hold automatically.

[mrbobbytables]

Superseded by #1245, which replaces the regex scanners with parser-based ones (the approach decided on #1194) and covers this fix, with regression tests. Closing in favour of that PR.