strict_syntax ^ (XOR) and relational sides crash check_equality/check_order instead of giving feedback #307

Description of the bug

With strict_syntax: true, ^ is kept as Python XOR, which is intended. But a response or answer that uses it crashes grading, and the learner gets an EvaluationError instead of feedback.

• check_equality: x^2 parses to a boolean (Not), and lhs_expr - rhs_expr raises TypeError: unsupported operand type(s) for -: 'Not' and 'Not'.
• check_order (app/context/symbolic/comparison.py): it computes lhs_expr - rhs_expr with no guard. Fix TypeError when the answer or response is an inequality #297 added a guard to check_equality only, so an order criterion with a boolean or relational side fails in the same way.
• x(x) is read as a call of a function x, so it is graded incorrect against x^2 with no explanation.

Expression(s)/input that trigger it

Expected vs. actual result

• Expected: ^ stays XOR. The response is graded incorrect, and the feedback includes NOTATION_WARNING_EXPONENT ("use ** for powers"). check_order treats a side it can't subtract as "cannot determine", as check_equality does.
• Actual: EvaluationError.

Steps to reproduce / where it was used

evaluation_function("x^2", "x^2", {"strict_syntax": True}). Same result on pre-refactor main, on main, and on the stack top.

Jira: LAM-292

🤖 Generated with Claude Code