fix symbolic field handling by CommonSubExprElim

[mmcloughlin]

Fix a problem with the treatment of symbolic field variables by common-subexpression-elimination.

When a common subexpression has been identified, it needs to place it in the statement list. It does this by placing it in the first position after all its dependencies have been assigned. The problem comes because a common subexpression that depends on a symbolic field won't ever find an assignment to the symbolic field.

This is what the algorithm was probably supposed to do, but as they explain in a comment what they actually do is place the definition of the common subexpression in the first place after all its dependencies have appeared (whether assigned to or not). Therefore in practice, common expressions involving symbolic fields would end up placed after their first use.

This PR fixes it by:

• Passing a list of globals into CSE which are treated as predefined
• Appending symbolic fields to the global identifier set