pdf: 0.5.1 — fix ref, 0.5.0 was pinned to a commit missing all its features

[asubbarao]

0.5.0 (#2218) was pinned to ref: 9006e93, which is the batch-1 baseline — it predates every feature the 0.5.0 description advertises. At that commit pdf_images, pdf_split_blank, pdf_watermark, pdf_bates, pdf_redact, and pdf_sign don't exist, so a user who installs 0.5.0 and calls any of them gets Catalog Error: function does not exist. My apologies for the bad ref in #2218 — the description was written for the feature surface but the ref pointed at the last commit that had cleared all-platform CI at the time.

This bumps the ref to 177b634 (duckdb-pdf main), where all of those functions are present and tested, plus:

• auto-OCR is now best-effort — a blank/scanned page encountered during a plain read_pdf no longer hard-errors on machines without a Tesseract model installed; it degrades to empty text for that page. Explicit ocr := true still raises the actionable install error. (Without this, the extension's own test suite failed on the registry's model-less Linux runner.)
• pdf_revisions no longer fatally invalidates the connection on a header-less input — an unescaped %PDF- in the error format string had been escalating a clean IOException into a fatal InternalException.
• Windows-portable test assertions — two feature tests asserted a forward-slash path (Windows glob expansion returns backslashes) and trimmed ' \\n\\f' but not the '\\r' in Windows line endings; both fixed (test-only). This is what an earlier ref (d33b1f3) still tripped on the windows_amd64/MSVC leg.

Only the pinned ref changed vs the merged 0.5.0; the description already documents these functions.

[carlopi]

Thanks!