Draw an SVG referenced by an <image> in the image's viewport - #547

Fixes #237.

An <image> that references an SVG document draws that document at its own size and ignores the width, height and preserveAspectRatio of the <image>. With the files attached to #237, circle.svg (199.5mm square) is drawn about 560 units wide instead of filling the 68.8 x 69.9 rectangle drawn under it, where Inkscape and browsers place it.

SVG 1.1 §5.7 says the referenced document is rendered into the viewport of the <image>: the image's width and height replace the document's, and the image's preserveAspectRatio replaces the document's own. renderSvg() already fits a nested <svg> to its viewport with preserveAspectRatio, so the new _render_svg_in_image() sets these three attributes on the referenced root <svg> while it is rendered, and restores them afterwards, because the same document can be referenced again at another size (ExternalSVG caches it). An <image> without width/height keeps the document's size, as before (tests/samples/others/svg_in_svg.svg is unchanged), and a reference to a fragment (file.svg#id) is not affected.

With the #237 files, the circles now fill the rectangle exactly (bounds 50.6, 49.8 to 119.5, 119.7, the rectangle's box).

Added test_svg_in_image_fills_the_image_viewport, which compares an <image> of an SVG file with the same content as an inline nested <svg>, for preserveAspectRatio="none", the default and xMinYMax slice, plus a second reference to the same file at another size; it fails on main. Updated tests/public_api.json with --update for the new private name, and added a ### Fixed entry in CHANGELOG.md (I used #546, I'll fix it if the PR gets another number). pytest (without the network-dependent sample downloads), ruff format --check, ruff check and mypy src tests pass.

AI-assisted (Claude Code); I reviewed and tested the change