Add native zero-inclusive symlog axis scale

[FarhanAliRaza]

Motivation

• Long-tailed numeric measures commonly include valid zeros which a log axis cannot represent while a linear axis compresses the tail, so a zero-inclusive long-tail scale (symmetric log / symlog) is needed.
• The goal is to provide a native axis type that preserves original data-space values for ticks, hover, selection, pan/zoom, density rendering, and payloads without leaking display transforms into application logic.

Description

• Add symlog as an accepted axis type_ and a constant parameter to x_axis()/y_axis() and Figure.set_axis() to configure the linear region width around zero, including validation of a positive constant. (python/xy/components.py, python/xy/_figure.py)
• Serialize scale: "symlog" and constant into axis payloads and ensure Figure._axis_spec emits scale/constant while preserving existing semantics for linear/log. (python/xy/_figure.py)
• Implement symmetric-log transforms in the static PNG/SVG pipeline via _Scale and tick generation that produce ticks in original data units (including zero). (python/xy/_svg.py)
• Wire symlog through the browser-side code: add symlog branch to axis mapping, implement JS coordinate transform and inverse, propagate a per-axis constant uniform into GLSL, and update shader map/calls so WebGL rendering, density, hit/pick, and interaction layers operate using the symlog transform consistently. (python/xy/static/index.js, python/xy/static/standalone.js)
• Add documentation for the zero-inclusive long-tail scale and an integration test exercising API, payloads, static transforms, and ticks. (docs/core-concepts/axes-and-scales.md, tests/test_symlog_axis.py)

Testing

• Ran focused and integrated Python tests: PYTHONPATH=python python -m pytest tests/test_symlog_axis.py -q (8 passed) and PYTHONPATH=python python -m pytest tests/test_figure.py tests/test_svg_export.py tests/test_png_export.py -q (212 passed, 1 skipped), and PYTHONPATH=python python -m pytest tests/test_components.py -q (76 passed).
• Static JavaScript validation succeeded with node --check python/xy/static/index.js and node --check python/xy/static/standalone.js and git diff --check reported no issues.
• Browser conformance / screenshot smoke could not be run here because Playwright’s Chromium executable is not installed in this environment (so end-to-end screenshot verification was not executed).

---

Closes #83

[codspeed-hq]

Merging this PR will not alter performance

✅ 94 untouched benchmarks
⏩ 1 skipped benchmark¹

---

_{Comparing codex/add-zero-inclusive-long-tail-axis-scale (7e6e73c) with main (b283b10)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩

[FarhanAliRaza]

without vs

with

[Alek99]

Review findings:

• [P1] js/src/40_gl.js:76-78: symlog is applied after decoding offset-encoded f32 coordinates. With x=[0, 1, 1e12] and constant=1, 0 and 1 collapse to the same decoded value even though symlog should visibly separate them. Encode geometry in scale coordinates or retain sufficient precision before the transform.
• [P1] js/src/40_gl.js:348: density aggregation remains uniform in raw data space. Clusters at x=1 and x=1000 can land in the same raw grid cell even though they are far apart in symlog display space. Bin/decimate in scale coordinates.
• [P2] python/xy/_svg.py:2260: static heatmap/density grids are stretched only between transformed endpoints, so internal cells are not symlog-transformed. PNG follows the same pattern.
• [P2] python/xy/_figure.py:993: scale="symlog" changes renderer semantics but the protocol remains 3; an older client will silently treat it as linear. Bump the protocol and regenerate bundles.