Decode AcroForm field names and values as PDF text strings

[litoosh13]

Fixes #504.

Form field names (/T) and text/choice values (/V) were decoded with String::from_utf8_lossy. They are PDF
text strings (PDF 32000-1 §7.9.2.2): UTF-16 with a byte-order mark, or PDFDocEncoding. This change uses the
crate's existing decode_pdf_text_string for the three sites in src/extractor/links.rs (the /URI and the
checkbox name stay as they are, as noted in the issue).

One consequence that #504 does not mention: the U+FFFD characters end up in the page's text, the page is then
flagged needs_ocr with suspected_garbled_text, and extract_pages_markdown returns empty markdown for the
whole page — the printed text included. A filled German form whose field names contain umlauts
(PDFDocEncoding, no UTF-16) loses all its text this way, so this is not only about Acrobat's UTF-16 values.

The test covers both encodings: a PDFDocEncoding name (Straße und Größe) and a UTF-16BE value
(Jürgen Groß). cargo test --lib: 1619 passed.

We have been carrying this fix in a patched copy of pdf-inspector in leafmind
(https://github.com/litoosh13/leafmind, third_party/pdf-inspector/leafmind-form-field-strings.patch).

---

Summary by cubic

Fixes #504 by decoding AcroForm field names (/T) and text/choice values (/V) as PDF text strings instead of reading them as UTF-8 with lossy replacement. This prevents non-ASCII names and values (PDFDocEncoding or UTF-16 with BOM) from becoming U+FFFD, which previously caused the page to be flagged as garbled and extract_pages_markdown to return empty markdown for the whole page. The test covers both encodings.

^{Written for commit d30f20b. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 1 file

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

<sub>Shadow auto-approve: would auto-approve. Replaces lossy UTF-8 decoding of AcroForm field names and values with the existing PDF text string decoder, fixing garbled non-ASCII form data; a new test covers PDFDocEncoding and UTF-16 values.

Re-trigger cubic</sub>