Fix weird valgrind transaction/connection issue

Once we were able to reproduce issue #6638 consistently on our local
machines we were able to use some manual guesswork at the causes and
`git bisect` to find the cause of it. Which turns out were two causes,
which together combined into the valgrind issue:
1. Transaction/connection cleanup refactoring in #6314
2. Adding replicationConnParam to the connection in #6080

The innocent looking `hash_uint32` call on the `replicationConnParam`
field (of type `bool`) that was added in #6080 is the main culprit. And
thus we address that here by replacing it with `hash_any`. It seems that
hash_uint32 was causing hashing of undefined memory somehow. But I'm not
entirely sure how. Since I'd expect the `bool` to be promoted to a `1`
of `uint32`. It might have something to do with the fact that
`hash_uint32` is defined as `inline`.

The transaction/connection cleanup refactoring (while afaict correct)
then caused this hashing issue to surface as a memory issue later on
during cleanup of the connection.

Both of these commits were included in 11.1, but my guess is that
valgrind didn't catch it then because #6314 was merged after the initial
release branch was cut.

Fixes #6638

src/backend/distributed/connection/connection_management.c

@@ -1237,7 +1237,8 @@ ConnectionHashHash(const void *key, Size keysize)
 	hash = hash_combine(hash, hash_uint32(entry->port));
 	hash = hash_combine(hash, string_hash(entry->user, NAMEDATALEN));
 	hash = hash_combine(hash, string_hash(entry->database, NAMEDATALEN));
-	hash = hash_combine(hash, hash_uint32(entry->replicationConnParam));
+	hash = hash_combine(hash, hash_any((void *) &entry->replicationConnParam,
+									   sizeof(entry->replicationConnParam)));
 
 	return hash;
 }