refactor(layout): extract the inline token model out of TextFlowSupport

[DemchaAV]

Why

TextFlowSupport is the largest file in document.layout (~1877 LOC). This is the first of two byte-identical steps that pull the inline paragraph wrapping subsystem out of it. Step 1 moves the leaf of that subsystem — the inline token model — so the follow-up wrapping-loop extraction has a clean dependency to build on.

What

• New package-private file InlineLayoutToken.java: the sealed InlineLayoutToken and its InlineTextToken / InlineImageToken / InlineShapeToken / InlineSvgToken record variants (including InlineSvgToken.toResolvedSvgLayer, the SVG-layer → engine-span lowering).
• The types were private nested in TextFlowSupport; as top-level package-private types the wrapping loops reference them by simple name unchanged (same package) — the modified file only loses code (−200: the token block + 6 now-unused imports).
• Pure type move, no logic edits, no public API change (document.layout is @Internal).

Tests

• Core suite (395) and the qa byte-identity gate (642 — InlineHighlightRenderTest, InlineSvgRenderTest, InlineShapeRenderTest, UnicodeFallbackDemoTest, TableCellInlineCodeWrapTest, PdfVisualRegressionTest, all *VisualParityTest, LayoutSnapshotRegressionDemoTest) green: renders are byte-identical.
• No new unit test: the move is a mechanical type relocation with zero behaviour change, guarded by the parity/visual suites above. A focused ParagraphWrappingTest lands with the wrapping-loop extraction (step 2).

Follow-up

Step 2 extracts the wrap loops + tokenize/trim/indent helpers + ParagraphIndentSpec into a ParagraphWrapping collaborator (rewiring the prepareParagraphLayout call sites).