Add parsed-document storage model to SDK and MCP

[suguanYang]

Summary

Introduces the shared parsed-document storage model consumed by the CLI, MCP, and Notebook, plus page-citation asset support.

SDK (src/)

• ParsedDocumentStorage interface + DiskParsedDocumentStorage reference impl (src/storage/): manifest / chunk pages / assets / resumable sync-progress, keyed by documentId + revisionKey (revisionKey = jobResultId ?? jobId).
• knowledge.withParsedStorage({storage, scheduler?, limits?}) — reads try configured storage first (fresh iff manifest.revisionKey === revisionKey), fall back to remote documents.listChunks, return the bounded result immediately, and schedule a background sync on a miss.
• readChunks (paged display + assetUrlPolicy:"none"|"durable"), grepChunks (single-document, with continuation cursor), getDocumentOutline.
• syncParsedDocument — resumable, bounded by limits, commits the manifest only after all pages are present.
• Durable asset hardening: with assetUrlPolicy:"durable" + configured storage, visible chunk asset URLs (primary + metadata.pageAssets[]) are written into storage and rewritten; on failure the URL is omitted rather than exposing a presigned Knowhere URL.
• importJobResult (replaces deprecated cacheJobResult), loadJobResult sync-to-storage, page-citation asset generation + Piscina-backed PDF page renderer.

MCP (packages/mcp/)

• Updated to delegate to readChunks/grepChunks/getDocumentOutline with storage-first, remote-fallback behavior; descriptions and tests updated.

Testing

• SDK unit tests cover storage round-trips, storage-hit vs. remote-fallback reads, non-blocking sync scheduling, paged readChunks + invalid param combinations, single-document bounded grep with truncation, and outline fallback.
• Verified live against a local Knowhere API: documents.listChunks snake→camel transform, remote read → scheduler fire → syncParsedDocument → storage-hit fast path, getDocumentOutline, and grepChunks.

🤖 Generated with Claude Code

---

Companion PRs (coordinated parsed-storage rollout):

• CLI: Ontos-AI/knowhere-cli#7
• Notebook: Adopt SDK parsed-document storage for chunk reads, sync, and chat assets knowhere-notebook#127