feat!: adopt mod-utils v6 (mod-llm v2)

[markdumay]

Summary

• Bump this module's own Go path to github.com/gethinode/mod-llm/v2 and the
  mod-utils requirement to v6 v6.0.1 (never v6.0.0), across go.mod,
  config.toml, exampleSite/hugo.toml, README.md, and
  exampleSite/content/docs/getting-started.md. Part of the mod-utils v6
  adoption wave: feat!: redesign argument and type system as v6 (core + shims, golden-tested) mod-utils#334 (see the v5-to-v6 migration notes in
  mod-utils' README).
• No argument-handling call sites to migrate. mod-llm has zero
  InitArgs.html/InitTypes.html (or Args.html) call sites — it has no
  shortcodes or partials that validate arguments. Its llms-components.json
  generator (layouts/index.llmscomponents.json) reads site.Data.structures
  (the raw _arguments.yml/_types.yml schema files mod-utils mounts)
  directly, never through the Args API, so this PR is a mechanical path bump
  plus dependency cleanup, not an Args-API migration.
• hugo mod tidy (run from exampleSite/, which uses Hugo workspace wiring —
  mod-llm.work + use ../) drops the now-stale mod-llm v1.3.5 and
  mod-utils/v5 v5.23.4 requires from exampleSite/go.mod/go.sum. It also
  prunes a pre-existing unused mod-mermaid/v3 entry — confirmed via a
  pristine-tree hugo mod tidy run to be unrelated to this change (already
  prunable before this PR).

Verification

• hugo mod graph (from exampleSite/): github.com/gethinode/mod-llm/v2 →
  github.com/gethinode/mod-utils/v6@v6.0.1+vendor. Zero mod-utils/v5 edges
  anywhere in the graph.
• pnpm build (hugo --gc --minify -s exampleSite): exit 0, 0 ERROR lines,
  both before and after this change.
• Engine-endstate note (same phenomenon as documented for the
  mod-fontawesome pilot, feat!: adopt mod-utils v6 (mod-fontawesome v6) mod-fontawesome#329): this module's
  exampleSite is unusually rich for wave 1 — it mounts the full Hinode +
  mod-blocks + mod-docs + mod-bootstrap-icons stack (via a Hugo workspace) so
  its own llms.txt/markdown/JSON generators have real content to crawl.
  Because this module's go.mod now requires mod-utils/v6, it wins the
  shared layouts/partials/utilities/ mount for the whole exampleSite,
  flipping every other mounted module's InitArgs/Args calls onto the v6
  engine too (those modules are still individually pinned to v5, unreleased
  on v6). This surfaces ~60 new WARN lines and reflows ~12 screenshot pages,
  entirely from other modules' partials (assets/hero-image.html,
  assets/table.html, assets/sidebar.html, mod-blocks component wrappers,
  etc.) — not from mod-llm's own code, which has no call sites. Four of the
  affected pages (table, navs-and-tabs, video, file) reproduce the
  exact mechanism the program register already documented and accepted from
  the pilot's engine-endstate run ("legacy InitTypes derived UDT doc-tables
  from the GLOBAL type, ignoring the structure's inline type override — v6
  honors the override"); the rest are the same "nested defaults now apply
  correctly" / "unsupported nested attribute now reported" behavior the
  mod-utils v5-to-v6 migration notes document. None are fixable in this repo
  (different modules own that code) and none are ERROR lines.
• mod-llm's own output is byte-parity clean. Compared the module's
  distinctive output (not just HTML, since it mostly emits llms.txt-style
  text) between a pristine pre-migration build and this branch's build:
  llms.txt is byte-identical; every generated */index.md page is
  byte-identical except the one file this PR intentionally edited
  (docs/getting-started/index.md, the install-path snippet); and
  llms-components.json is structurally identical once the inherently
  volatile meta.generated build timestamp is normalized (all 60 components,
  all types, byte-for-byte).
• Visual harness (node tests/visual/visual.mjs) run against this module's
  own exampleSite, baseline shot from the pristine pre-migration clone
  before any edit: 61/73 pages pixel-clean (0.0000%–0.08%, under
  threshold), 12 flagged — all traced to the engine-endstate mechanism above
  via the build's own WARN output, none are unexplained.

Test plan

• pnpm build exits 0 with 0 ERROR lines (before and after)
• hugo mod graph (exampleSite context) shows mod-utils/v6 v6.0.1,
  zero mod-utils/v5 edges
• Confirmed zero InitArgs/InitTypes/Args.html call sites exist in
  this module (nothing to migrate)
• llms.txt / */index.md / llms-components.json byte-diffed against
  the pristine pre-migration build (see Verification above)
• Visual regression harness run against this module's own exampleSite,
  baseline captured before any change; all 12 flagged pages traced to a
  named, already-documented v6 engine-endstate mechanism in other
  modules' code
• Whole-repo grep for stale mod-llm (bare path) / mod-utils/v5
  references: zero hits after this change
• Commit-body landmine check
  (git show -s --format=%B HEAD | grep -E '^-.*-$'): no match

🤖 Generated with Claude Code

[markdumay]

🎉 This PR is included in version 2.0.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀