C#: Use CFG nodes instead of AST nodes in sign/modulus analysis

[hvitved]

This PR rewrites sign/modulus analysis for C# to use CFG nodes instead of AST nodes, in order to take CFG splitting into account. For example, in

void Splitting2(bool b)
{
    int x;
    if (b) x = 1; else x = -1;

    System.Console.WriteLine(x); // strictly positive (b = true) or strictly negative (b = false)

    if (b)
        System.Console.WriteLine(x); // strictly positive
    else
        System.Console.WriteLine(x); // strictly negative
}

x exists in two copies in the CFG at the first System.Console.WriteLine(x), and the new analysis is able to deduce that it is strictly positive in the copy for b = true and strictly negative in the copy for b = false.

The implementation uses (a copy of) the ControlFlowReachability library that we also use to take CFG splitting into account in the data-flow library.

I had to make two changes in the shared interfaces:

• rankedPhiInput is now moved out into the language-specific modules, as we need to rank on two columns for C#. It turned out that the old implementation was partial, which resulted in diverging analysis for loops at the beginning of a callable. The new (total) implementation is based on a similar pattern from the basic block library.
• I added a getSourceType predicate to CastExpr.

[aschackmull]

Java changes LGTM.

[tamasvajk]

LGTM, we'll see with RangeAnalysis if I understood everything in the PR. :-)