Update coding standards for constexpr if statements; NFC

We currently suggest that users not use an else clause after a return
statement in a prior if branch. e.g.,

if (foo)
  return 1;
else // Should remove this else clause
  return 10;

however, this suggestion is incorrect for a constexpr if statement
because one of the two branches will be a discarded statement and thus
can impact template instantiation behavior. This updates the coding
standard to make it clear that it's okay to have a return after an else
in a constexpr if statement.

I think this is an NFC change to the intent of the rule, which is why
I've not started an RFC for the changes.

Differential Revision: https://reviews.llvm.org/D132232

llvm/docs/CodingStandards.rst

@@ -1026,6 +1026,24 @@ Or better yet (in this case) as:
 The idea is to reduce indentation and the amount of code you have to keep track
 of when reading the code.
 
+Note: this advice does not apply to a ``constexpr if`` statement. The
+substatement of the ``else`` clause may be a discarded statement, so removing
+the ``else`` can cause unexpected template instantiations. Thus, the following
+example is correct:
+
+.. code-block:: c++
+
+  template<typename T>
+  static constexpr bool VarTempl = true;
+
+  template<typename T>
+  int func() {
+    if constexpr (VarTempl<T>)
+      return 1;
+    else
+      static_assert(!VarTempl<T>);
+  }
+
 Turn Predicate Loops into Predicate Functions
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^