C#: Add a hash-cons library for C#

[MathiasVP]

This PR adds a hash-cons library for C# as requested by @ropwareJB.

The goal of a hash-consing library is to create an efficient way to compare two expressions without creating a quadratically sized relation. For example, if you were to check if two VarAccesses are "syntatically equal" you may create this relation:

predicate syntacticallySimilar(VarAccess va1, VarAccess va2) {
  va1.getTarget() = va2.getTarget()
}

However, this relation would be of size n^2 where n is the number of accesses of a variable in the program. For example, if you have this program:

Use(x); // 1
Use(x); // 2
Use(x); // 3

this relation would include:

• A row in syntacticallySimilar that relates x on line 1 to x on line 1
• A row in syntacticallySimilar that relates x on line 1 to x on line 2
• A row in syntacticallySimilar that relates x on line 1 to x on line 3
• A row in syntacticallySimilar that relates x on line 2 to x on line 1
• A row in syntacticallySimilar that relates x on line 2 to x on line 2
• A row in syntacticallySimilar that relates x on line 2 to x on line 3
• A row in syntacticallySimilar that relates x on line 3 to x on line 1
• A row in syntacticallySimilar that relates x on line 3 to x on line 2
• A row in syntacticallySimilar that relates x on line 3 to x on line 3

i.e., exactly 3^2 = 9 rows!

To avoid this, hash-consing creates a "unique representation" of the structure of an expression. Crucially, an expression is mapped to exactly 1 hash-cons value. And if two expressions map to the same hash-cons value then they have the same structure. You can then compare whether two expressions are syntactically identical by checking if their hash-cons values are equal. This avoids the quadratic blowup.

I'm sure I have missed a number of expressions 😅 This will result in the library not realizing that two expressions are syntactically similar (because they will each be assigned a hash-cons unique to that expressions and thus they will never be equal to anything but that expression).

To gauge which expressions are most important (and haven't yet been handled by the library) you can run a query such as:

from string s, int k
where
  k =
    strictcount(Expr e | hashCons(e) instanceof TUniqueHashCons and s = e.getPrimaryQlClasses() | e)
select s, k

this will output how many expressions we couldn't map to a "useful" hash-cons value for each kind of expression. This is what I've used to "guide" myself on which expressions should be handled in this first PR, but I'm sure there are DBs where the numbers are totally different

[MathiasVP]

CI is unhappy because we merged #185 (which is necessary for the 2.21.0 upgrade, but it won't compile until then)