Refactor (slice 1, #242): extract diagnostics projection into DiagnosticsCollector

[chubes4]

Refs #242 — slice 1 of the HtmlTransformer decomposition epic: extract the diagnostics projection concern.

What this does

Pure extraction/move. The inline block in HtmlTransformer::transform() that projected already-computed conversion signals into the flat result diagnostics array is moved into a dedicated, cohesive module:

src/HtmlToBlocks/Diagnostics/DiagnosticsCollector.php

HtmlTransformer now delegates to DiagnosticsCollector::collect() with the data it already computed. What moved:

• the html_to_blocks_core_slice base diagnostic
• html_static_script_metadata projection (from scriptMetadata)
• fallback → diagnostic projection (html_*_fallback, interactive_control_behavior_lost, etc., with full finding enrichment fields: pattern_family, source_selector, parent/ancestor_reason, suggested_generic_repair_class, …)
• preserved_runtime_island projection (from runtimeIslands)
• wp_block_validity_* findings projection (source Runtime::class)
• html_semantic_parity_* findings projection

FallbackDiagnostic is untouched (the new module orchestrates the data it produces). No result schema/key changes. self::class attribution is preserved by passing the transformer source string into the collector; Runtime::class attribution is preserved by importing Runtime.

Behavior-preserving — parity identical 128→128

Captured a baseline before refactoring and re-ran after:

• composer test:canonical: 5 contracts pass (no-WP runtime, WP-stub runtime, plugin bootstrap, HTML-to-blocks, format bridge scaffold) — identical before/after.
• composer parity: 128 fixtures pass → 128 fixtures pass — identical.
• composer test (canonical + parity + packaging): green.

No logic changes, no new findings, no renamed codes, no fixture changes. Output is byte-identical.

Line-count reduction

HtmlTransformer.php: 6816 → 6723 lines (−93), with the projection logic now isolated in a 148-line single-responsibility module that diagnostics work can target without touching transformer core.

Scope notes

The semantic-parity report builders and the interactive-control behavior-loss / capture*Fallback emitters were intentionally left in place for this slice: they are deeply interwoven with shared DOM/conversion helpers (attr, elementSelector, safeFallbackHtml, boundedFallbackHtml, and normalizedNavigationLabel which is also used by navigation de-duplication), so moving them would not be a clean pure move. This slice extracts the largest provably-safe cohesive unit — the projection seam — leaving those emitters for a later slice once their shared helpers are factored out.

Do not merge without review.

🤖 Generated with Claude Code