Python: Fix bad join in PointsToContext.appliesTo

Two related fixes. First, I removed the pragmas from `appliesTo`, as
these were actually causing a huge cartesian product to be created in
`RangeIterationVariableFact.pointsTo` and
`ReModulePointToExtension.pointsTo`. On `python/cpython`, at least, this
caused a reduction of tuples from ~2.7 billion to ~2 billion.

However this also caused `PointsToContext.fromCall/2` to get a worse
join-order, so I inlined the definition and added a pragma _there_, and
that fixed the issue.

Author: tausbn

python/ql/lib/semmle/python/pointsto/PointsToContext.qll

@@ -149,7 +149,7 @@ class PointsToContext extends TPointsToContext {
 
   /** Holds if `call` is the call-site from which this context was entered and `outer` is the caller's context. */
   predicate fromCall(CallNode call, PointsToContext caller) {
-    caller.appliesTo(call) and
+    caller.appliesToScope(pragma[only_bind_into](call.getScope())) and
     this = TCallContext(call, caller, _)
   }
 
@@ -187,11 +187,7 @@ class PointsToContext extends TPointsToContext {
 
   /** Holds if this context can apply to the CFG node `n`. */
   pragma[inline]
-  predicate appliesTo(ControlFlowNode n) {
-    exists(Scope s |
-      this.appliesToScope(pragma[only_bind_into](s)) and pragma[only_bind_into](s) = n.getScope()
-    )
-  }
+  predicate appliesTo(ControlFlowNode n) { this.appliesToScope(n.getScope()) }
 
   /** Holds if this context is a call context. */
   predicate isCall() { this = TCallContext(_, _, _) }