fix(tools): PDFページングを損失なしカーソル方式へ再設計

PDF本文をページ単位で抽出した後に文字数上限で切断しつつ、継続位置だけ次ページへ進めていたため、ページ末尾が復元不能になる問題を修正する。ツール側の本文上限とLLM側の結果上限の不整合、25ページの処理上限、添付経路での継続情報欠落も共通契約で解消する。

- 共通PageResultと、メタデータ込みの結果上限内で本文量を決めるフォーマッターを追加

- v2トークンにPDFのページ番号とページ内文字位置を保持し、v1トークンの読み取り互換性を維持

- pdf_to_text、scrape_url、search_and_fetchを返却済み本文だけカーソルが進む方式へ統一

- max_pdf_pagesを文書全体ではなく1呼び出しの処理上限として扱い、長い単一ページと26ページ目以降の継続取得に対応

- attachment:<file_id>をセッション限定で解決し、既存添付も保存済みプレビューではなく元PDFから再解析

- Web取得ファイルをセッションとURL単位で重複排除し、HTMLおよびテキスト経路の全文ページングも維持

- 正常なページング結果の後段切断を禁止し、結果上限・コンテキスト予算・ラウンド上限では未取得状態を明示

- 日英プロンプト、UIメッセージ、ツールスキーマ、変更履歴を実装へ同期

- 実PDF、30ページ超、長い単一ページ、旧トークン、セッション分離、コンテキスト境界の回帰テストを追加

検証:

- pytest: 130 passed

- ruff check src tests: passed

- mypy src: passed

- c2144.pdf: 19ページ、抽出本文46015文字（ページ区切り込み46051文字）を8回で重複・欠落なく完全再構成

1 files changed, 42 insertions, 28 deletions

diff --git a/src/ollama_web/tools/scrape.py b/src/ollama_web/tools/scrape.py
index 39dab83..2f85c38 100644
--- a/src/ollama_web/tools/scrape.py
+++ b/src/ollama_web/tools/scrape.py
@@ -2,13 +2,37 @@
 
 from __future__ import annotations
 
+import logging
+
 from bs4 import BeautifulSoup
 
 from ..config import settings
-from .helper.pagination import format_paged_result, make_token, parse_token
+from .helper.pagination import (
+    PageSizeError,
+    format_page_to_limit,
+    make_token,
+    parse_token,
+    text_page,
+)
 from .helper.safe_http import UnsafeURL, fetch_public_bytes
 
 _TOOL_NAME = "scrape_url"
+logger = logging.getLogger("ollama_web.tools.scrape")
+
+
+def extract_readable_text(data: bytes, content_type: str = "") -> str:
+    """Decode and normalize the complete readable body of an HTML payload."""
+    encoding = "utf-8"
+    if "charset=" in content_type:
+        encoding = content_type.rsplit("charset=", 1)[-1].split(";", 1)[0].strip()
+    html = data.decode(encoding or "utf-8", errors="replace")
+    soup = BeautifulSoup(html, "html.parser")
+    for tag in soup(["script", "style", "noscript", "template", "head", "nav", "footer", "aside"]):
+        tag.decompose()
+    container = soup.find("article") or soup.find("main") or soup
+    text = container.get_text(separator="\n")
+    lines = [line.strip() for line in text.splitlines()]
+    return "\n".join(line for line in lines if line)
 
 
 def scrape_url(url: str, max_chars: int | None = None, continue_from: str | None = None) -> str:
@@ -47,43 +71,33 @@ def scrape_url(url: str, max_chars: int | None = None, continue_from: str | None
         # The token's URL takes precedence; it identifies the page being paged.
         url = str(payload.get("url") or url)
         offset = int(payload.get("offset", 0))
-        if offset < 0:
-            offset = 0
 
     try:
         data, final_url, content_type = fetch_public_bytes(url, timeout=settings.scrape_timeout)
-        encoding = "utf-8"
-        if "charset=" in content_type:
-            encoding = content_type.rsplit("charset=", 1)[-1].split(";", 1)[0].strip()
-        html = data.decode(encoding or "utf-8", errors="replace")
         url = final_url
     except UnsafeURL as exc:
         return f"Blocked URL: {exc}"
     except Exception as exc:  # noqa: BLE001
         return f"Failed to fetch {url}: {exc}"
 
-    soup = BeautifulSoup(html, "html.parser")
-
-    # Remove non-content tags.
-    for tag in soup(["script", "style", "noscript", "template", "head", "nav", "footer", "aside"]):
-        tag.decompose()
-
-    # Prefer article/main content when available.
-    container = soup.find("article") or soup.find("main") or soup
-    text = container.get_text(separator="\n")
-    # Collapse blank lines and trailing whitespace per line.
-    lines = [ln.strip() for ln in text.splitlines()]
-    cleaned = "\n".join(ln for ln in lines if ln)
+    cleaned = extract_readable_text(data, content_type)
 
     if not cleaned:
         return "No readable text content found."
 
-    # Slice the requested page from the full cleaned text.
-    page = cleaned[offset : offset + limit]
-    if not page:
-        # Offset is past the end; nothing left to return.
-        return format_paged_result("", has_more=False, next_token=None)
-
-    has_more = len(cleaned) > offset + limit
-    next_token = make_token(_TOOL_NAME, url, offset + limit) if has_more else None
-    return format_paged_result(page, has_more=has_more, next_token=next_token)
+    try:
+        result = format_page_to_limit(
+            lambda body_budget: text_page(cleaned, offset, body_budget),
+            lambda cursor: make_token(_TOOL_NAME, url, cursor["offset"]),
+            requested_chars=limit,
+            result_limit=settings.max_tool_result_chars,
+        )
+        logger.info(
+            "scrape page offset=%d result_chars=%d has_more=%s",
+            offset,
+            len(result),
+            "---page-info---" in result,
+        )
+        return result
+    except (PageSizeError, ValueError) as exc:
+        return f"Scrape pagination failed: {exc}"