[Effects] Simplify handling of LookupExpr in effects checking

Our logic for doing the "declaration reference" classification was
unnecessarily convoluted, and did "unsafe" classification twice for
properties and subscripts that have other effects (throws/async) on
their getters. Simplify it.

Author: DougGregor

lib/Sema/TypeCheckEffects.cpp

@@ -1398,7 +1398,8 @@ class ApplyClassifier {
     if (isa<SubscriptDecl>(cdr.getDecl()))
       return PotentialEffectReason::forSubscriptAccess();
 
-    assert(isa<VarDecl>(cdr.getDecl()));
+    // Note that declarations other than properties end up with
+    // "property access" effect reasons for historical... reasons.
     return PotentialEffectReason::forPropertyAccess();
   }
 
@@ -1537,18 +1538,18 @@ class ApplyClassifier {
     if (!member)
       return Classification::forInvalidCode();
 
-    PotentialEffectReason reason =
-        PotentialEffectReason::forPropertyAccess();
-    Classification classification;
+    ConcreteDeclRef declToClassify;
     if (auto getter = getEffectfulGetOnlyAccessor(member)) {
-      reason = getKindOfEffectfulProp(member);
-      classification = Classification::forDeclRef(
-          getter, ConditionalEffectKind::Always, reason, E->getLoc(),
-          assumedSafeArguments && assumedSafeArguments->contains(E));
-    } else if (isa<SubscriptExpr>(E) || isa<DynamicSubscriptExpr>(E)) {
-      reason = PotentialEffectReason::forSubscriptAccess();
+      declToClassify = getter;
+    } else {
+      declToClassify = member;
     }
 
+    PotentialEffectReason reason = getKindOfEffectfulProp(member);
+    Classification classification = Classification::forDeclRef(
+          declToClassify, ConditionalEffectKind::Always, reason, E->getLoc(),
+          assumedSafeArguments && assumedSafeArguments->contains(E));
+
     classification.setDowngradeToWarning(
         downgradeAsyncAccessToWarning(member.getDecl()));
 
@@ -1557,14 +1558,6 @@ class ApplyClassifier {
         E->isImplicitlyAsync().has_value(), E->isImplicitlyThrows(),
         reason);
 
-    if (!classification.hasUnsafe()) {
-      classification.merge(
-          Classification::forDeclRef(
-            member, ConditionalEffectKind::Always, reason, E->getLoc(),
-            assumedSafeArguments && assumedSafeArguments->contains(E),
-            EffectKind::Unsafe));
-    }
-
     if (assumedSafeArguments) {
       // If the declaration we're referencing is explicitly @safe, mark arguments
       // as potentially being safe.