Fix #637: recognise rho-walking references as void-attr usages

[bibonix]

@yegor256

Closes #637.

Why

The unused-void-attr lint flagged a void attribute as unused even when it was clearly referenced from inside a nested formation. For example:

# Foo.
[widths] > tabular
  [i] > inner
    widths.at i > @

It warned that widths is unused, but widths.at i inside the inner [i] formation is a usage — it just walks back through one ρ hop to reach tabular.

The XSL only checked for the literal substring ξ.<attr> in @base, so any reference parsed as ξ.ρ.<attr>, ξ.ρ.ρ.<attr>, etc. slipped past it.

How

• src/main/resources/org/eolang/lints/misc/unused-void-attr.xsl — match ^ξ(.ρ)*.<attr>(...)?$, count the ρ hops, and confirm that the formation declaring the attribute is exactly that many enclosing abstract scopes above the usage. References that walk past the formation, or that resolve into a different scope, still leave the attribute flagged as unused.
• src/test/resources/.../allows-void-attr-used-in-nested-formation.yaml — new regression test covering the bug from the issue (widths.at i from inside a nested formation).
• src/test/resources/.../allows-usage-as-nested-rho-ref.yaml — the previously-named catches-usage-as-nested-rho-ref pack was locking in the buggy behaviour, so it is renamed and its assertion flipped: ^.x from a nested formation is a real usage of the outer x.

All the existing shadowing tests (catches-unused-void-attribute-with-same-named-formation, catches-unused-void-attribute-horizontally-with-same-name, catches-void-with-same-name-in-canonical, catches-unused-void-attr) keep passing, because rho-walking is structural rather than name-based: an inner formation that declares its own x as void does not absorb the outer scope’s ξ.ρ.ρ.x.

How verified

• mvn -B test -Dtest=LtByXslTest#testsAllLintsByEo — 345 packs pass, 1 skipped (unchanged).
• Manual run of Source(...).defects(...) on the snippet from False positive unused-void-attr warning for used attribute #637: no unused-void-attr warning is emitted any more.

Note about CI

The mvn (ubuntu/macos/windows), deep, and reserved jobs are red on this PR, but they have been red on every master commit since 2026-04-26 (last green run #2202 on 2026-04-24). I confirmed locally that the same two tests time out on plain master too: MonoLintsTest.lintsProgramCorrectly and PkByXslTest.doesNotDuplicateDefectsWhenMultipleDefectsOnTheSameLine. Those failures are not introduced by this change.