Add .value() method for expression evaluation and refactor variable retrieval

[haozhu10015]

Summary

This PR closes #1.

• Adds Expr::value(&solution) to evaluate any expression — variables, sub-expressions, or composite atoms — at an
  optimal solution
• Adds Index<(usize, usize)> for Array to enable natural result[(row, col)] indexing on returned values
• Updates all examples to use .value() for variable retrieval; least_squares demonstrates both &solution[&var] and var.value(&solution) as equivalent APIs
• Adds expression evaluation demos where sub-expressions carry meaningful insight: fitted values in least squares, actual return in portfolio, fit/regularization split in LASSO

[haozhu10015]

While writing the .value() tests I noticed that sum_squares(Ax + c) in the objective was silently returning the wrong answer when the expression had a nonzero offset or a non-identity coefficient matrix. The old canonicalization fell through to an SOC path that minimized ‖Ax + c‖ instead of ‖Ax + c‖^2, so the reported objective was always the square root of the correct value.

A fix is ready on feat/sum-squares-fix (rebased on top of this branch). Happy to bundle it here or keep it as a follow-up PR.

[SteveDiamond]

Thanks for fixing!

[SteveDiamond]

Let's do the second fix as a follow up.

[SteveDiamond]

Actually could you get the CI passing? Some minor issues.

[haozhu10015]

The lint is now fixed but the MSRV error is because the latest zmij and also quote does not support rust 1.70, shall we bump the MSRV to 1.71 or cap the zmij version (< 1.0.21)?

[SteveDiamond]

The lint is now fixed but the MSRV error is because the latest zmij does not support rust 1.70, shall we bump the MSRV to 1.71 or cap the zmij version (< 1.0.21)?

Lets bump MSRV to 1.71