fix: stop applying PX_TO_PT to font-size twice (#502)

[Zunhammer]

The viewport group already carries the px-to-pt conversion, but font-size was converted again below it: applyStyleOnShape mapped it through convertLengthToPt, and convertText measured advance widths in points before adding them to x coordinates in user units. Text therefore rendered at 0.75x the size of the geometry around it, and multi-tspan runs were laid out in the wrong unit. Both now stay in user units, leaving the single conversion to the viewport transform.

Three tests in tests/test_units.py asserted the stored String.fontSize in points and so encoded the bug; they now assert the rendered size instead. The relational tests there and in test_basic.py are untouched and still pass. Adds regression coverage for the text-to-geometry ratio across seven root variants, for multi-tspan positioning and for per-character positioning; all six fail without the source change.

[github-actions]

Thank you for making your first pull request! Your contribution to svglib is highly appreciated!

[deeplook]

Thanks for the detailed writeup and repro — this is a genuine regression from the 2.0b1 unit rework, not expected behaviour like #471.

Verified independently: your repro script now emits Tf 16 inside the .75 viewport matrix (matching 1.5.1 output, vs. the buggy Tf 12), full test suite passes locally (188 passed, 5 skipped), and all 16 CI checks are green.

Fix is well-scoped, the rewritten tests correctly assert rendered size instead of raw String.fontSize, and the changelog entry is clear about the unit-semantics change. Merging — thanks for catching this.