chore(release): prepare 0.59.0 on develop

[lmeyerov]

Summary

Prepare pygraphistry 0.59.0 on develop after the GFQL benchmark/docs stack landed.

• Merge current master@76dc3f302 ancestry into current develop@d32f83afa so the later develop → master promotion is conflict-free.
• Combine every unreleased changelog entry from both branches and cut ## [0.59.0 - 2026-08-31].
• Preserve the landed docs(gfql): sync benchmark contract v3 #2011 benchmark prose and fix(gfql): preserve Polars across graph CALL #2012 Polars CALL boundary tests.
• Keep package versioning tag-derived; there is no source version literal change.
• Give the Python 3.12 Polars coverage cell deterministic margin after exact-head CI proved its fully passing 2-worker run exceeded the old whole-job budget.

Diff shape

The PR is broad because master contains 125 files of already-landed/reviewed changes that develop does not yet contain. Those blobs are inherited from master, not newly authored here. The manual release work is limited to the changelog merge/release cut, the reviewed q1–q9 merge-resolution spelling, and the narrow Polars CI-budget correction. The final sync from develop adds only #2012's reviewed test amplification.

The 0.59.0 changelog contains the exact parent union: 100 entries on develop, 130 on master, and 133 unique inherited entries, with none missing or duplicated. It adds one candidate-authored Infrastructure receipt for the Polars timeout correction.

Exact head: 749736ca8aa3219185ad6a3acee37449aeb230e1

Binary diff fingerprint against develop: 93a1e6d99bd711aa13d6d6c01925a866e6118c17294baa8d0997cbeb8b9337a4

Validation

• 48 focused benchmark-doc, Polars CALL boundary, and schema-effect tests passed.
• Repository-native full Ruff, type-hygiene, comment-density, and relative-import gates passed.
• Repository-native focused production mypy passed; exact-head CI owns the full pinned type matrix.
• Docs LaTeX Unicode, credential-pattern, residue, shell syntax, and git diff --check gates passed.
• A temporary unpushed 0.59.0 tag built the expected universal wheel and sdist; wheel metadata reports graphistry==0.59.0. The tag and all build outputs were deleted.
• The original Polars CI receipt had 7,091 tests pass in 566.14 s, its append phase pass 129 tests in 13.89 s, and its coverage audit pass before the 10-minute whole-job cancellation. The budget is now 15 minutes; test selection, workers, and coverage floors are unchanged.
• No Docker or heavy local compute was used.

Landing and release boundary

This PR targets develop. After maintainer review/merge, open a clean develop → master promotion PR for maintainer landing. Do not tag or publish until that promotion is on master and the separate pre-tag readiness checkpoint is approved.