fix: symbolic sign, Max/Min, and numeric copysign in the SymPy backend

[henryiii]

🤖 AI text below 🤖

Part of #711.

Fixes the SymPy backend's _lib shim and related issues:

sign raised on symbolic input

_lib.sign delegated to numpy.sign, which cannot evaluate a symbolic relational. Repro (before):

>>> r, a = sympy.Symbol("r", positive=True), sympy.Symbol("a", real=True)
>>> vector.VectorSympy2D(rho=r, phi=sympy.Symbol("phi", real=True)).scale(a)
TypeError: cannot determine truth value of Relational: a < 0

Now uses sympy.sign for symbolic input (numpy for plain numbers).

maximum/minimum returned the first symbolic argument unconditionally

The two methods were byte-identical, so symbolic clamps were silently dropped — e.g. deltaangle's acos domain clamp and the t2/Mt2 floors. They now return sympy.Max/sympy.Min for symbolic input, and fall back to numpy.maximum/numpy.minimum for plain numbers (so nan from mixed sympy/object operations propagates instead of raising inside sympy.Max).

The exact-expression assertions in the sympy mirror tests were updated to the new honest forms (e.g. acos(Min(1, ...)), sqrt(Max(0, tau**2 + ...))). No numeric .subs(...).evalf() expectation changed.

copysign — documented, deliberately not fully fixed

copysign(val1, val2) returning val1 assumes the sign carrier is non-negative (true for timelike vectors). I attempted the faithful Abs(val1) * sign(val2) (and a Piecewise variant matching numpy's copysign(x, 0) == +x): both are mathematically right, but tau/t round-trips then produce nested Piecewise expressions that crash sympy's piecewise_fold/simplify_logic with ValueError: The argument 'nan' is not comparable, and degrade every tau-coordinate expression with unsimplifiable sign(...) factors. The shortcut is now explicitly commented (with a pointer to #711), and the numeric path (copysign(3.0, -2.0)) is handled correctly via numpy.copysign. A proper symbolic encoding of signed tau is left as a follow-up decision.

Other

• Added the missing BSD-3 copyright headers to src/vector/backends/sympy.py and src/vector/_pytree.py.
• Fixed an operator-precedence bug in two _wrap_result branch conditions (the isinstance/issubclass guards bound to only one arm of the or; latent today, aligned with the correct parenthesization in numpy.py).
• New regression tests: symbolic scale, symbolic Max/Min, numeric copysign.

Tests: full suite 814 passed, 3 skipped (unrelated optional deps); prek -a clean.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.17%. Comparing base (d9ef1db) to head (1ef85a9).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #721   +/-   ##
=======================================
  Coverage   87.16%   87.17%           
=======================================
  Files          96       96           
  Lines       11199    11205    +6     
=======================================
+ Hits         9762     9768    +6     
  Misses       1437     1437           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.