Add SIPG elliptic operator with inhomogeneous Dirichlet/Neumann BC support and convergence coverage

[Copilot]

This PR adds an interior-penalty DG elliptic model implementing the SIPG bilinear form shown in the issue (volume diffusion term plus symmetric consistency and penalty face terms). It also adds manufactured-solution convergence coverage on a square for both inhomogeneous Dirichlet and Neumann boundary data.

• New elliptic model (grudge/models/elliptic.py)

  • Adds InteriorPenaltyEllipticOperator.
  • Supports configurable diffusion coefficient kappa and penalty scaling via penalty_factor.
  • Implements interior-face flux assembly with jump/average structure and ( \alpha_e / h_e ) penalty scaling.
  • Supports inhomogeneous boundary data:
    • dirichlet_tag, dirichlet_bc
    • neumann_tag, neumann_bc
  • Accepts scalar, field, or callable BC inputs.

• Model export

  • Re-exports InteriorPenaltyEllipticOperator from grudge/models/__init__.py for direct import from grudge.models.

• Convergence tests (test/test_elliptic.py)

  • Adds h-convergence test with a quadratic manufactured solution on a square mesh.
  • Covers both boundary-condition modes:
    • all-Dirichlet
    • all-Neumann
  • Uses the operator residual against the known source term to verify expected convergence behavior across mesh refinement.

from grudge.models.elliptic import InteriorPenaltyEllipticOperator
from meshmode.mesh import BTAG_ALL, BTAG_NONE

elliptic = InteriorPenaltyEllipticOperator(
    dcoll,
    kappa=1.0,
    penalty_factor=10.0,
    dirichlet_tag=BTAG_ALL,
    dirichlet_bc=u_boundary,   # scalar / DOFArray / callable
    neumann_tag=BTAG_NONE,
)

residual = elliptic.operator(u) - f