feat(svm): draw a curve as a curve, not as a line through its control points

[andiwand]

🤖 Generated with Claude Code

Stage 9 of #772, closing the last of its ten defects, stacked on #790 →
#789 → #788 → #787 → #786 → #785 → #784 → #779 — review those first; this
branch's base is feat/svm-map-mode.

A polygon that carries curves is written twice: first its corners alone,
then again as Polygon::Read writes it — VersionCompat, the points, and one
PolyFlags per point. We read the flag byte that says the second one follows,
dropped it, and drew the first. Every curve therefore came out as a line
through what were its control points.

PolyFlags::Control marks a control point, three in a row are one bezier
segment, and that is an svg C. A poly-polygon says the same thing through its
"complex polygons": each names by index the polygon it replaces with the same
shape, curves included — also read and dropped until now.

The polygon a shape is made of therefore carries its flags now, which is what
svgwriter.cxx's GetPathString walks too.

Checked against LibreOffice

For a metafile written by hand with one flagged polygon, LibreOffice's own
export writes

M 100,700 C 300,100 700,100 900,700

and so do we, character for character.

Nothing in the corpus moves

Not one of the 1125 metafiles harvested from the fixtures carries a polygon
flag or a complex poly-polygon, so no reference output changes and there is no
pin bump. The 3 new tests (47 in the svm suite) are inline bytes: a polyline
with flags curves, the same polyline without them is lines, and a complex
poly-polygon replaces its polygon.

What is left after this

svm/PLAN.md names it: WALLPAPER, FLOATTRANSPARENT and EPS, which all
nest something of their own; the MASK family; MOVECLIPREGION;
ZCOMPRESSed bitmaps; and version-1 (pre-VCLMTF) files. None of them occurs
in the corpus, and each is logged where it is skipped.