Add warning when symbolic info is lost in the interpreter.

[BenSpencer]

Some operations and combinations of parameter types are not supported in the symbolic interpreter, so the result of these operations on symbolic inputs is still concrete (#119 is an example).

We should add warnings in all cases where we "knowingly" drop symbolic information in the interpreter.

[BenSpencer]

Added statistics in 09aeaf0. There were only two of these cases in the symbolic interpreter itself.

I am not sure what to do about ail_call and ail_call_native. They do nothing in the symbolic interpreter and I am not clear if we are losing information here.

[BenSpencer]

• Native calls are already either instrumented, give warnings or intentionally ignored (and listed in docs as such).
• JS calls are implemented by the interpreter, so there is no need to "handle" any symbolic info in the call instructions themselves.

So I don't believe ail_call and ail_call_native have any problem.

However, symbolic information may still be missed in a few other corner cases. For example, an array or object equality check where one array has a symbolic element. In these cases we don't see the whole array as symbolic (as they are not supported symbolically) and therefore do not see that a symbolic check for a[5] == b[5] with a[5] symbolic has "secretly" taken place.

I am not sure how to handle/detect these cases.