fix(lint): deduplicate CSS transform text in gsap_css_transform_conflict finding

[santhiprakash]

Problem

In the gsap_css_transform_conflict lint rule, when a single CSS
transform declaration contains both translate and scale functions
(e.g. transform: scale(1.08) translate3d(1.5%, 0, 0)), the rule
stores the same transformVal in both cssTranslateSelectors and
cssScaleSelectors for the same selector. The finding message and
fixHint then concatenate the identical string via
[cssFromTranslate, cssFromScale].filter(Boolean).join(" "), producing
doubled text like:

".scene-1" has CSS transform: scale(1.08) translate3d(1.5%, 0, 0) scale(1.08) translate3d(1.5%, 0, 0)

instead of the correct:

".scene-1" has CSS transform: scale(1.08) translate3d(1.5%, 0, 0)

Root cause: gsap.ts line ~1356 — the join has no deduplication,
so when both maps resolve to the same value for the same selector, it
appears twice.

Triage / Root cause

packages/lint/src/rules/gsap.ts — gsap_css_transform_conflict rule
handler. The two regex tests (/translate/i and /scale/i) both match
the same combined declaration, so both cssTranslateSelectors and
cssScaleSelectors store the identical transformVal. The downstream
.filter(Boolean).join(" ") then doubles it.

Fix

• Deduplicate [cssFromTranslate, cssFromScale] via new Set() before
  joining into cssTransform.
• This is a no-op when the two values are different (separate
  translate-only and scale-only CSS declarations for the same selector)
  and corrects the doubling when they are identical.
• Added a regression test to the existing "combined CSS transform
  conflicts with multiple GSAP properties" test case that asserts the
  message and fixHint do not contain the doubled transform text.

Verification

bun test packages/lint/src/rules/gsap.test.ts

162 tests pass (0 fail). The new assertion confirms no doubling:

expect(msg).not.toContain("translateX(-50%) scale(0.8) translateX(-50%) scale(0.8)");
expect(hint).not.toContain("translateX(-50%) scale(0.8) translateX(-50%) scale(0.8)");

Notes / Risks

• Minimal change — only affects the string construction path, not the
  conflict detection logic itself.
• No existing behavior changes for the common case where translate-only
  and scale-only CSS rules are separate (the new Set() is a no-op
  when parts contain different values).

Closes #3263