C#/Java: Materialize sink/source/summary predicates to avoid bad join order.

[michaelnebel]

Running java/telemetry/supported-external-api-sinks on glopezGitHub/Android23 (database artifact found here: https://github.com/codeql-testing/glopezGitHub--Android23/actions/runs/4050012179)
yielded a log containing:

[2023-02-07 09:36:48] Evaluated non-recursive predicate _Expr#20226873::Call::getCallee#0#dispred#ff_ExternalApi#3b305853::ExternalApi::getAnInput#0#dispred__#count_range@3483fcrm in 644206ms (size: 102815).
Evaluated relational algebra for predicate _Expr#20226873::Call::getCallee#0#dispred#ff_ExternalApi#3b305853::ExternalApi::getAnInput#0#dispred__#count_range@3483fcrm with tuple counts:
             357828        ~0%    {3} r1 = SCAN Expr#20226873::Call::getCallee#0#dispred#ff OUTPUT In.1, 1, In.0
             357828        ~0%    {3} r2 = r1 AND NOT _Expr#20226873::Call::getCallee#0#dispred#ff_GeneratedFiles#c1ec4513::GeneratedFile#class#f_Location__#antijoin_rhs(1, Lhs.2, Lhs.0)
             357828        ~0%    {3} r3 = JOIN r2 WITH Member#9eba3c33::Callable::getSourceDeclaration#0#dispred#ff ON FIRST 1 OUTPUT Rhs.1, 1, Lhs.2
        24250368013        ~2%    {4} r4 = JOIN r3 WITH ExternalApi#3b305853::ExternalApi::getAnInput#0#dispred#ff ON FIRST 1 OUTPUT Rhs.1, 1, Lhs.2, Lhs.0
         5820107998  ~5344227%    {3} r5 = JOIN r4 WITH ExternalFlow#030a639a::Cached::sinkNode#2#ff ON FIRST 1 OUTPUT Lhs.3, 1, Lhs.2
         1649881354  ~1631058%    {4} r6 = JOIN r5 WITH ExternalApi#3b305853::ExternalApi::getApiName#0#dispred#bf ON FIRST 1 OUTPUT Rhs.1, Lhs.2, Lhs.0, 1
                                  return r6

To avoid the join to getAnInput before the filtering (sinkNode) we materialize the predicate isSink (that is, add pragma[nomagic]). We do the same for isSource and hasSummary as they could potentially exhibit the same issue. The same is done for C#.

It turned out that after introducing the pragmas on the predicates mentioned above another bad join was introduced in the characteristic predicate for TestLibrary where the name of all elements were matched against the regexp before joining only with those elements that represented packages. This has also been fixed.

[michaelnebel]

DCA looks good.