Expose bounded PDF bytes on EmptyPdf without refetching #104

Problem

WebClaw v0.6.19 already buffers the complete response body in Response.body, then calls webclaw_pdf::extract_pdf. For scanned/image-only PDFs, PdfError::EmptyPdf is returned and the already-fetched bytes are dropped. A caller that needs an OCR or vision fallback cannot recover those exact bytes; calling fetch_raw would perform a second source request and can observe different content.

Requested narrow seam

Please expose an opt-in outcome for the final response already buffered by fetch_and_extract_with_options when PDF parsing returns EmptyPdf:

• original PDF bytes from that response;
• final URL;
• content type;
• byte length;
• SHA-256;
• stable reason such as empty_pdf.

Existing extraction methods and ordinary CLI output should remain unchanged. A new API can return an enum such as Extracted(ExtractionResult) | Artifact(PdfArtifact). If exposed by CLI, an explicit flag plus JSON/base64 output would keep it machine-readable; the caller should supply a strict artifact-size ceiling, and bytes must not appear in logs or error formatting.

Acceptance

• one source fetch for the returned artifact; no fetch_raw retry;
• artifact bytes are exactly the bytes passed to extract_pdf;
• only the opt-in path can return bytes;
• over-limit artifacts fail closed;
• final URL/content type/size/SHA-256 are tested;
• existing callers and output remain backward compatible.

This does not ask WebClaw to add OCR or a model dependency. It only preserves a bounded handoff for callers that already own that policy. Is this API shape acceptable upstream?