Fix MAX_ELLIPSOID crash for low-dimensional (1D/2D) polytopes

[Akash504-ai]

This PR fixes a crash that occurs when computing the MAX_ELLIPSOID rounding
for low-dimensional (1D and 2D) polytopes.

Background

The MAX_ELLIPSOID rounding procedure computes the minimum and maximum
eigenvalues of the ellipsoid matrix in order to evaluate the roundness
condition. For performance reasons, this is done using
Spectra::SymEigsSolver, requesting the smallest and largest eigenvalues.

However, Spectra enforces the constraint:
1 <= nev <= d - 1
where d is the matrix dimension.

The previous implementation always requested nev = 2, which is valid for
d >= 3 but invalid for d <= 2. For 2D polytopes (d = 2), this results in
nev > d - 1 and causes Spectra to throw a std::invalid_argument exception,
terminating the program.

Fix

This PR updates the eigenvalue computation logic as follows:

• For d > 2, the existing Spectra-based partial eigen decomposition is used
  unchanged.
• For d <= 2, the code falls back to Eigen::SelfAdjointEigenSolver to compute
  the full spectrum safely.

This avoids invalid Spectra calls and ensures that MAX_ELLIPSOID works
correctly for low-dimensional polytopes, while preserving the original
behavior and performance characteristics for higher dimensions.

Notes

The crash is deterministic for d <= 2 and does not depend on input data.
Although the lp_solve-dependent example could not be run locally due to
environment constraints, the fix removes the exact failure path responsible
for the exception.

Fixes #325.

[Akash504-ai]

Closing temporarily to reduce PR load as discussed. Will reopen sequentially once the current PRs are reviewed/merged.