RFC: Refactor rendering API and native font pipeline for v1.0.0

[behackl]

Summary

This PR replaces the existing rendering interface with a smaller, explicit v1 API
and consolidates rendering around immutable results, SVG-space geometry, and
renderer-owned font state.

I don't intend to merge this branch, just started development at v0 at some point
and did not want to rebase against main (which has two previous 1.0.0 pre-releases
already) -- but should probably merged into / replace main while creating a v1
release branch. At that point I'd probably also change the default branch of this repo.

It is probably quite obvious, but I want to be explicit about it: this has been cooked up
with extensive usage of LLMs; I've kept (coarsely) reviewing individual commits and
in particular the test suite though. Not included: the integration of the new interface
in Manim -- but it effectively warrants a complete rewrite of Text and friends like
Paragraph to get rid of all the old, ugly and effectively unmaintainable code. I strongly
believe that the current redesign allows a much, much cleaner implementation of
multi-line text in particular within Manim (while at the same time, or so I hope at least,
forming a more maintainable interface here). Curious to hear what you think!

The main public entry points in this refactored version are:

• render() for literal plain text
• render_markup() for Pango markup
• TextSpan for structured plain-text styling
• RenderedText, Bounds, and LineInfo for SVG output and layout metadata
• register_font() / FontRegistration for explicit custom-font lifetimes
• get_version_info() for linked-library version information

Design highlights

• Plain text and markup are deliberately separate paths; plain text is never
  interpreted as markup.
• All public geometry, including font sizes, uses SVG user-space units.
• Render results and spans are immutable value objects.
• Font registration is reference-counted, renderer-owned, and isolated from
  unrelated Linux Fontconfig application state.
• Markup is parsed once per render and shared across measurement and SVG output.
• Span normalization and line metadata handling now scale linearly or near
  linearly for large styled or multiline inputs.
• Native error handling and cleanup paths were tightened, including Cairo status
  checks and explicit rejection of NUL-containing native inputs.

Infrastructure and documentation

• CI, wheel validation, release triggering, packaging metadata, and docs were
  modernized alongside the API work.
• Documentation includes the migration and breaking-change overview; this PR
  description intentionally stays focused on the implementation shape.

Verification

• Full test suite, docs/doctests, linting, wheel smoke tests, and the current
  cross-platform CI matrix are green.

[behackl]

To ease into this patchbomb (sorry!), I can recommend starting with the explicit changelog page in the documentation at https://manimpango--129.org.readthedocs.build/en/129/changelog.html, followed by reading test output.

This also modernises the CI/CD pipeline, the actual checks are currently not reported on this PR -- but you should be able to see them over at my fork: https://github.com/behackl/ManimPango/actions.

[naveen521kk]

I'll take a look this weekend

[Hun-Bot2]

I tested this rewrite on macOS against the glyph-position rounding issue discussed in #130 and ManimCommunity/manim#2844.

The new rendering path still reproduces the same behavior. For example:

render("AVATAR", font="Arial", size=10.0)
advances: 6.0  6.0  6.0  5.0  7.0

At very small sizes, llllllllll at size 1.667 collapses all glyph positions to the same x-coordinate in my test.

One subtlety: the positions have a fractional origin offset (0.015625 in my environment), so checking absolute x positions for fractional values can hide the issue; the pairwise advances remain integer-quantized.

Since this rewrite already requires Pango >= 1.44, the round_glyph_positions fix proposed in #130 should be applicable here without the compatibility guard needed by v0.

Leaving this here so the issue can be accounted for in the v1 rendering path.