fix: ignore comment + whitespace nodes in template-no-yield-only

[johanrd]

Cowritten by claude:

Summary

• template-no-yield-only flags yield-only templates via templateNodes.length === 1 && isYieldOnly(templateNodes[0]). That holds only while the parser strips template comments out of the body, which [email protected] happened to do.
• Upstream [email protected] (NullVoxPopuli/ember-estree#31) started keeping MustacheCommentStatement nodes in the body (the native Glimmer AST shape). Under that parser a template like <template>{{! x }}{{yield}}</template> has a body of length 2 and is silently no longer flagged. This is the rule failure that blocks Update ember-eslint-parser to 0.11 #2735's CI.
• Fix: filter comment + whitespace-only text nodes before the length check, via the same isEmptyNode helper that upstream ember-template-lint and the sibling rule template-no-bare-yield.js already use.

Why fix it here rather than in the parser / estree

[email protected]'s change is the correct AST: it matches the pure-hbs parser (ember-eslint-parser/hbs) and the native Glimmer AST, both of which have always kept comment nodes in the template body. Re-hiding those nodes at the parser layer would re-introduce the bug that #2733 guards against ({{! eslint-disable-* }} directives would again be invisible to tooling that needs to see them). The semantic decision "should a comment-plus-yield template count as yield-only?" belongs to the rule.

What this does not fix

Latent bug still remain after this PR (candidates for follow-up)

1. {{yield to="inverse"}} is still wrongly flagged. The initial Extract rule: template-no-yield-only #2596 commit had (!node.hash || !node.hash.pairs || node.hash.pairs.length === 0) guarding against this — the AI correctly called it "an
   improvement" over upstream. The sync commit dropped it. A template whose only body is {{yield to="inverse"}} is not truly "yield-only" — it's explicitly delegating to the else / named <:inverse> block, which is
   the idiomatic way to do that. Upstream has this bug; we inherited it on the sync commit; we could diverge better.

separate follow-up?

Notes

• template-no-bare-yield.js and template-no-yield-only.js are near-duplicate ports of ember-template-lint/lib/rules/no-yield-only.js; only the former had kept pace with upstream's empty-node filter. This closes that drift.
• The filter is a no-op on the currently-pinned [email protected], so this is safe to land independently of Update ember-eslint-parser to 0.11 #2735 and paves the way for that PR's CI to go green once rebased.
• This also fixes a latent bug the rule had on 0.10: <!-- html -->{{yield}} was not being flagged as yield-only; now it is.

Test plan

• pnpm test tests/lib/rules/template-no-yield-only.js — 22 passed (18 original + 4 new assertions: {{!-- long --}}{{yield}} and <!-- x -->{{yield}} against both the gjs and hbs parsers)
• npx prettier --check + npx eslint clean on both modified files
• CI green