bannertopdf: support Unicode text with embedded fallback fonts

[oubayashi]

Summary

Fix garbled non-ASCII text in bannertopdf output.

Banner fields were previously written as raw bytes using the PDF standard
Courier font, which only works reliably for single-byte text. Printer names,
locations, job titles, user names, and other fields containing UTF-8 text
could therefore appear garbled or lose characters.

This pull request also fixes an existing page-resource handling issue exposed
while testing Unicode font embedding. Banner templates with an indirect
/Resources dictionary could lose their image XObjects when banner fonts were
added.

Changes

Unicode banner text

• Add strict UTF-8 decoding with validation for malformed, overlong, surrogate,
  and out-of-range sequences.
• Use bit masking to validate UTF-8 leading and continuation byte patterns.
• Use Fontconfig to find scalable outline fonts covering non-ASCII characters.
• Embed selected fonts as Unicode CID fonts through PDFio.
• Support per-character font fallback when a line contains multiple writing
  systems.
• Cache and reuse up to 32 embedded fonts.
• Keep the existing Courier path for ASCII text.
• Escape parentheses, backslashes, and control characters in PDF literal
  strings.
• Replace unsupported characters safely and emit warnings instead of producing
  invalid font references.
• Apply the Unicode path consistently to printer information, job metadata,
  user names, host names, billing information, model names, and localized
  times.

Preserve template resources

• Detect whether the page /Resources entry is a direct dictionary or an
  indirect object.
• Resolve and copy indirect resource dictionaries before adding banner fonts.
• Preserve existing /XObject, /ExtGState, /Font, and other template
  resources.
• Prevent images from default-testpage.pdf from disappearing when the page
  resources are updated.

Regression Test

Add an end-to-end regression test following the existing
test-pdftopdf-inherited-mediabox test pattern.

The test runs a mixed ASCII and Unicode banner through the existing
testfilters harness, then reopens the generated PDF with PDFio and verifies
that:

• The output is a valid single-page PDF.
• The original Courier resource remains available for ASCII text.
• A Unicode Type 0 fallback font resource is created.
• The Unicode font contains a ToUnicode map.
• The fallback font file is embedded in the PDF.
• Unicode text is emitted using the embedded font.
• Parentheses and backslashes are correctly escaped in PDF literal strings.
• The template image resources /x6, /x7, and /x8 are preserved.
• The corresponding /x6 Do, /x7 Do, and /x8 Do drawing commands remain
  in the page content stream.

The testfilters harness now supports selecting bannertopdf explicitly for
focused regression testing.

The regression test is skipped when Fontconfig, PDFio, or a font covering the
test character is unavailable.

The initial regression-test implementation was prepared with assistance from
OpenAI Codex and reviewed against the existing project test patterns.

The changes were built and tested successfully on Ubuntu 24.04. Testing
included generating a banner with mixed ASCII and Unicode text, confirming
that the fallback font was embedded, and verifying that the text and template
images were rendered correctly in the generated PDF.

Compatibility

• Pure ASCII output keeps the existing behavior and layout.
• Unicode font fallback is enabled only when Fontconfig is available.
• Without Fontconfig, unsupported non-ASCII characters fall back to ?.
• Supplementary-plane characters above U+FFFF currently fall back to
  U+FFFD, because PDFio 1.6 uses a BMP-sized CID-to-GID mapping.
• Complex-script shaping and bidirectional text are outside the scope of this
  change.

Commit Structure

This pull request is split into two commits:

1. Add UTF-8 decoding, Fontconfig fallback, embedded Unicode CID fonts, PDF
   string escaping, and an end-to-end Unicode banner regression test.
2. Preserve indirect page resource dictionaries when adding banner fonts and
   extend the regression test to cover template image resources.

[oubayashi]

I added a separate follow-up commit for the missing template images found
during testing.

The default test-page template stores /Resources as an indirect object.
The existing font setup path treated it as missing and replaced it with a new
dictionary, dropping the template's image XObjects. The follow-up commit now
copies and preserves indirect resources before adding banner fonts.

The regression test has also been extended to verify the /x6, /x7, and
/x8 image resources and their drawing commands.