feynlag 0.2.0

Added

• diagonalize_takagi(method="auto"|"symbolic"|"numeric", dps=50): for a
  numeric matrix larger than 2×2, auto factorises with mpmath's eigsy at
  high precision (columns by increasing mass, round-off eigenvalues as exact
  zero modes), so a generic 3+2 seesaw 5×5 finishes instead of stalling in the
  exact route. Exact 2×2 and symbolic calls are unchanged.
• Per-symbol LaTeX: component_tex=[...] on Scalar, WeylFermion/
  MajoranaFermion, GaugeBoson and GaugeGroup.bosons(), tex= on
  conjugate_pair, and a parameter's existing tex= now reach plain
  sympy.latex through the symbol itself (TexSymbol), so downstream code
  needs no name→LaTeX map. A one-component field's tex= also names its
  component, and fermion bar legs print as \overline{…}.
• Opt-in LaTeX names in the SM builders and Scalar.expand_vev: each takes
  tex=, a map from the name of a symbol it creates to a LaTeX name
  (electroweak_gauge, higgs_doublet, electroweak_scaffold,
  weinberg_rotation, charged_current_rotation, to_physical_basis,
  standard_ckm), so one global table covers the couplings, the Higgs
  parameters, the doublet and its fluctuations (H0_r→h), and the physical
  Z, A, W±, G⁻. Names missing from the map stay plain. An explicit
  symbol_names printer setting still overrides a symbol's own tex.
  GaugeGroup.bosons() raises if a later call asks for a different tex than
  the cached bosons carry.
• A TexSymbol sorts exactly like a plain Symbol of the same name, so
  adding a tex never reorders vertex legs or printed terms.

Fixed

• The UFO writer now escapes texname/antitexname. It wrote them as bare
  '...' literals, so a tex containing a quote (g') made parameters.py
  a SyntaxError, and a backslash was read as a Python escape (\tau →
  TAB + au).

Changed

• A field component or parameter declared with a tex is a TexSymbol, which
  is not equal to a plain Symbol of the same name. Code that rebuilds such a
  symbol by name (sp.Symbol("lam")) must use the object instead. Without a
  tex nothing changes.