feat(xlsx): honour header/footer font and size codes

[developer0hye]

Summary

headerFooter/oddHeader and oddFooter strings carry &"Font,Style" and &<n>
codes selecting the face and size for the runs after them.
parse_hf_format_string parsed both and threw them away — the branches were
literally "skip to closing quote" and "skip digits" — so every header and footer
printed in the fallback serif.

Measured

tests/fixtures/xlsx/headerFooterTest.xlsx, whose six slots each set
&"Calibri,Regular". Embedded fonts:

	fonts in the output
before	LibertinusSerif-Regular + Carlito-Regular
after	Carlito-Regular alone

The serif is gone: header and footer text now takes Calibri, which resolves to
Carlito here. Cell text was already Carlito, so the serif was the header/footer
and nothing else.

Change

Each section becomes a Vec<HfSegment> rather than one String. A font or size
code opens a new segment carrying the updated style, so a code partway through a
section applies from that point on rather than to the whole slot —
&Lplain&"Calibri,Regular"styled yields two runs, only the second with a face.

&"-,Bold" keeps the current face and sets the style; a style word replaces both
flags, so Excel's Regular turns them back off. bold/italic stay None when
no style word appears, leaving the renderer's own default rather than pinning
false.

Scope

The &<n> size code is parsed and applied, but this fixture sets no size
code, so its 11pt-vs-12pt difference is not addressed here. The issue reports
that as part of the same defect; measured, it is separate — Excel's default
header size, not a discarded code — and changing a default on this evidence
would be guessing. The code path is covered by a unit test instead.

Blast radius

3 of 69 XLSX fixtures change: headerFooterTest, date, date_1904 — the ones
with font codes in their header strings.

A correction to earlier PRs in this series while I am here: my local main build
had been failing since #790 (a gitignored Cargo.lock pinning a docx-rs
commit without word_wrap), and cargo build's error was going into a line I
discarded, so target/release stayed stale. Blast-radius counts I reported for
#793 through #797 were measured against that stale binary and are inflated —
this PR's own sweep initially read 5 of 69 and drops to 3 once main is built
correctly, the two extras being #795/#796's icon changes misattributed here.
Every targeted before/after in those PRs was checked against a committed GT, so
those numbers stand; the sweep counts do not.

Tests

Four. One caught a real bug: open_segment cloned the previous segment to
inherit an unchanged face and brought its text along, duplicating the earlier
run's text into the new one.

Visual impact

• No rendered PDF change

• Rendered PDF change or visual evidence added

• Reason: the change is a font family selection, and this environment has
  neither Calibri nor the GT's Arial installed — both resolve to metric-
  compatible substitutes (Carlito, Liberation Sans), so a rendered comparison
  here would show the substitution rather than the fix. The embedded-font list
  above is the evidence: the fallback serif is no longer used for header text.
  tests/fixtures/xlsx/headerFooterTest.xlsx has no committed GT.

Related: #633