feat(kebab-chunk): P7-2 pdf-page-v1 chunker — page-aware splitting

`PdfPageV1Chunker` 가 `kebab-parse-pdf` 가 emit 한
`CanonicalDocument` (블록당 한 페이지, 모두 `SourceSpan::Page`) 를 받아
페이지 경계를 절대 넘지 않는 `Chunk` 들을 생성. `chunker_version =
"pdf-page-v1"`.

핵심 동작:
- 페이지 텍스트가 `target_tokens × BYTES_PER_TOKEN` (= 3) 안이면 한
  덩어리. 초과 시 `\n\n` (paragraph) 또는 sentence-end 구두점 + whitespace
  경계를 segment 로 보고 greedy 누적, 기본 한 chunk 당 최소 한 segment.
- 다음 chunk 의 prefix 에 `overlap_tokens × BYTES_PER_TOKEN` 만큼의 직전
  꼬리를 prepend (char 단위, 이전 chunk 시작 너머로 backtrack 안 함).
- 빈/공백-only 페이지는 0 chunk (페이지의 `Provenance::Warning` 으로
  `kebab-parse-pdf` 단계에 이미 표시됨).
- 비-PDF doc (Block::Paragraph 가 아니거나 SourceSpan 이 Page 아님) →
  명시 에러.

Spec deviation (HOTFIXES 2026-05-02 P7-2):
- `chunk_id` 충돌 가드: 같은 페이지에서 여러 chunk 가 나오면 `block_ids`
  가 모두 같아 §4.2 recipe 가 충돌. `id_for_chunk` 의 `policy_hash` 인풋을
  per-chunk 로 `format!("{base}#c{char_start}")` 변형해 회피. recipe 자체는
  불변. `Chunk.policy_hash` 필드는 base 유지.
- `BYTES_PER_TOKEN = 3` (md-heading-v1 실제 코드와 일치). spec 본문은
  "/ 4" 라고 했지만 그 자체가 md-heading-v1 의 실코드와 어긋나 있어 일관성
  쪽을 택함. cross-chunker `policy_hash` 동일성 unit test 로 잠금.

테스트 (10개 신규):
- chunker_version label, 3-page small, 1-page huge + overlap + chunk_id
  유일성, empty page skip, whitespace-only skip, non-PDF error,
  cross-page boundary 절대 안 만들어짐, determinism (1000회), snapshot
  shape 안정, md-heading-v1 와 policy_hash 동일.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tasks/HOTFIXES.md

@@ -14,6 +14,26 @@ historical contract that was implemented; this file accumulates the
 deltas so phase 5+ readers can find the live behavior without diffing
 git history.
 
+## 2026-05-02 — P7-2 pdf-page-v1: chunk_id collision + BYTES_PER_TOKEN
+
+**Discovered**: P7-2 implementation start.
+
+**Symptom 1 (load-bearing)**: `tasks/p7/p7-2-pdf-page-chunker.md` § Behavior contract literally says `chunk_id` per design §4.2 with `(doc_id, "pdf-page-v1", block_ids, policy_hash)`. But unlike `md-heading-v1` (which always emits at most one chunk per atomic block), `pdf-page-v1` splits one page-block into multiple chunks when page text exceeds the byte budget. All sub-chunks of the same page have identical `block_ids` → identical `chunk_id` collisions, breaking the §3.5 invariant that `chunk_id` is a primary key.
+
+**Symptom 2 (cosmetic)**: Spec text says `token_estimate = byte_len / 4` and "matches `md-heading-v1` proxy". Looking at the actual md-heading-v1 source (`crates/kebab-chunk/src/md_heading_v1.rs:17`), the constant is `BYTES_PER_TOKEN = 3` (chosen to cover Korean ≈ 3 b/tok and over-estimate English ≈ 4 b/tok). Spec's "/4" claim is inconsistent with the implementation it claims to match.
+
+**Root cause**: §4.2 chunk_id recipe was designed assuming one-chunk-per-block-set. Page-aware chunking violates that assumption.
+
+**Fix** (PR #38, feat/p7-2-pdf-page-chunker):
+
+- **Per-chunk policy_hash variant**: feed `format!("{base_policy_hash}#c{char_start}")` into `id_for_chunk`'s `policy_hash` slot so chunks within the same page get distinct `chunk_id`s. The §4.2 recipe itself stays unchanged — only the *input* to one of its slots differs per chunk. The unmodified `base_policy_hash` is still stored in `Chunk.policy_hash` so the field still answers "what policy was active" (workspace-wide policy invalidation lookups continue to work).
+- **`BYTES_PER_TOKEN = 3`** (matches md-heading-v1 actual code, not spec literal). Cross-chunker policy fingerprint identity is verified by a unit test: `policy_hash_matches_md_heading_v1_for_identical_policy`.
+
+**Trust note**: The per-chunk hash variant is opaque (`#c<n>` is just a marker, not interpretable as char_start by downstream tools — they read `Chunk.source_spans[0].char_start` for that). Downstream identifier comparisons on `chunk_id` continue to work as opaque blake3 hashes.
+
+**Amends**:
+- tasks/p7/p7-2-pdf-page-chunker.md (chunk_id recipe per-chunk variant; BYTES_PER_TOKEN = 3 not 4).
+
 ## 2026-05-02 — P6-3 caption: GenerateRequest.images + cargo feature dropped
 
 **Discovered**: P6-3 implementation start.