
Specifically, the CPython interpreter's symbol visibility analysis makes
recursive calls into a function called `symtable_visit_expr`, and can
overflow the stack when analyzing a very large expression of binary
operators (this also came up in this ticket:
https://trac.sagemath.org/ticket/16014). While I seek a more general
fix to that problem, we need to work around it. In this case the
polynomial `psi` is converted one term at a time, and then summed,
rather than summed first and then converted (resulting in evaluating a
very large Singular polynomial). The result is the same either way
though might be slightly slower with this workaround, though I haven't
tested.
