fix(core): stop scoring generic XML feed media types

[k2kirov]

Summary

• machine-discovery/feed-entry-identity-and-canonical-integrity no longer returns warn for a feed served as application/xml or text/xml. The warn cost half of a grade-B scored audit (0.3 weight) with no consumer source behind it.
  • Every feed reader parses a generic XML type, and the dossier names no consumer that treats it worse.
  • application/rss+xml was never registered with IANA, so the old message ("the registered rss media type") was false. Only Atom's application/atom+xml is registered (RFC 4287).
  • The type is still recorded under details.warnings, so the result schema is unchanged. A non-XML type such as text/html still fails.
• AGENTS.md (symlinked as CLAUDE.md): the changeset bump rule now follows the output contract instead of "any verdict moved":
  • major: an id is removed or renamed, a tier or grade moves, the schema changes, or a flag, tool or export is removed or changes meaning
  • minor: the contract grows
  • patch: a verdict is corrected to match its existing dossier, with id, tier, grade, weight and schema unchanged

Found on https://dwithease.com, which serves /rss.xml as application/xml.

Changeset

@forkpoint/agent-lighthouse-core: patch, under the new rule.

Test plan

• pnpm build && AL_SKIP_NETWORK=1 pnpm test && pnpm typecheck && pnpm lint && pnpm check:dossiers && pnpm check:requires && pnpm check:audit-map
• Test covers text/xml and application/xml (pass, type noted) and text/html (fail)
• CI runs the live-site suite