ci(gfx11): install Node.js so llama-ui builds from source

[jimw567]

Summary

The gfx11 ROCm build started failing at the Provisioning UI assets step
(llama-ui-embed failed (1) -> ninja: build stopped) after the recent
upstream sync (PR #43) pulled in the new SvelteKit web UI and its
scripts/ui-assets.cmake provisioning (upstream ggml-org#22937 moved the prebuilt UI
out of the repo and onto a HuggingFace bucket; ggml-org#23352 refactored the cmake
build).

ui-assets.cmake provisions the embedded web UI in three tiers:

1. pre-built assets checked into tools/ui/dist/ (not present),
2. npm build from source (what upstream CI uses),
3. HuggingFace bucket download (no-npm fallback).

The gfx11 runner has no npm (UI: npm not found, skipping npm build), so it
fell through to tier 3. There it 404'd on the pinned b1 bucket (the fork has
no real build number), fell back to the latest bucket, and that archive is
out of sync with the checked-out source: it extracted fine (cmake only checks
index.html) but llama-ui-embed rejected it for a missing required asset
(loading.html), which is fatal.

Fix

Add an actions/setup-node@v6 step (node 24) before the build, matching what
upstream's own workflows (release.yml, ui-build.yml) already use. With npm
on PATH the build takes tier 2 and compiles the UI from the checked-out source,
so the assets always match llama-ui-embed and there is no dependency on the
HF bucket being in sync. This mirrors upstream behavior and self-heals across
future syncs.

Test plan

• gfx11 CI build-ubuntu job reaches and passes the Provisioning UI assets step
• Build completes and artifacts upload