pdf_io: write the formatted buffer in pdf_dev_begin_actualtext

[peisbacher]

pdf_dev_begin_actualtext formats each character into the local buf, but hands
work_buffer to dev_out. That buffer is never written in this function, so the
/ActualText string in the output consists of whatever work_buffer happens to
hold, in practice a run of NUL bytes.

Upstream texk/dvipdfm-x/pdfdev.c in TeX Live passes buf here. The divergence
came in with 3f16760 ("crates/pdf_io: update for TeXLive 2021 xdvipdfmx"), which
rewrote several pdf_doc_add_page_content(work_buffer, len) calls into
dev_out(p, buf, len) and changed only the sprintf target in this one.

Reproducing

\documentclass{article}
\usepackage{fontspec}
\setmainfont{Fira Sans OT}
\XeTeXgenerateactualtext=1
\begin{document}
Philipp
\end{document}

The seven characters produce seven NUL bytes:

tectonic 0.17.0:   /Span << /ActualText (\0\0\0\0\0\0\0) >> BDC
xelatex (TL 2022): /Span << /ActualText (Philipp) >> BDC

Why it is worth fixing

/ActualText takes precedence over the font's ToUnicode CMap in poppler, so a
document built with \XeTeXgenerateactualtext=1 has no usable text layer at all:
pdftotext returns NUL bytes for every glyph in the affected spans. Copy-paste,
search, screen readers and CV/resume parsers all see nothing. Extractors that
ignore /ActualText, such as pypdf, are unaffected, which makes this easy to miss.

The flag exists precisely to correct a wrong ToUnicode mapping, so the documents
that set it are the ones that can least afford a broken text layer. In my case the
font maps U+002D, U+00AD, U+2010 and U+2011 to one glyph and xdvipdfmx picks the
highest codepoint for the reverse mapping, which put U+2011 where hyphens belonged.

Verification

Built this branch on macOS 27 (aarch64) and ran the snippet above:

release 0.17.0:  /Span << /ActualText (\0\0\0\0\0\0\0) >> BDC   pdftotext: \0\0\0\0\0\0\0
this branch:     /Span << /ActualText (Philipp) >> BDC          pdftotext: Philipp

All three encoding paths of the function check out:

input	/ActualText bytes	pdftotext
Philipp	Philipp	Philipp
Größe Öl Straße	Gr\xf6\xdfe, \xd6l, Stra\xdfe (PDFDocEncoding)	Größe Öl Straße
C‑Level (U+2011)	\xfe\xff\000C \021\000L... (UTF-16BE)	C‑Level

Happy to add a regression test if you would like one, though I did not see an
existing pattern for asserting on content-stream output.