Fix enumtables' table detection (#59)

### Description
enumtables uses an empty select to detect the presence of tables.  Unfortunately, as postgres wraps this whole process in a transaction, this can lead to situations where the tables are not detected properly.

Here's how it goes wrong:
1. It gathers all the tables that are enums.
2. It tests for the existence of the tables.  The first table check that fails will cause all subsequent operations in the transaction to fail, whether or not the operation is actually doable.
3. A lot of spurious table changes are "detected".

This PR uses the table detection logic alembic internally uses (refactored for our needs), and this produces a changeset that makes sense.

### Test plan
Autogenerated a schema migration with multiple enumtables.  It produced the migration script.

third-party/sqlalchemy-enum-tables/enumtables/alembic_autogen.py

@@ -1,5 +1,6 @@
 
 import alembic.autogenerate
+from alembic.autogenerate.render import _ident
 from sqlalchemy import MetaData
 
 from . import enum_column
@@ -13,16 +14,10 @@ def get_declared_enums(metadatas, schema, default):
 		for column in table.columns if (isinstance(column.type, enum_column.EnumType) and table.schema == schema))
 	return {typ.__enum__.__tablename__ : frozenset(typ.__enum__.__enum__.__members__) for typ in types}
 
-def is_table_present(fully_qualified_table_name, connection):
-	try:
-		connection.execute("SELECT * FROM {} LIMIT 0;".format(fully_qualified_table_name))
-	except:
-		return False
-	else:
-		return True
-
 @alembic.autogenerate.comparators.dispatch_for("schema")
 def compare_enums(autogen_context, upgrade_ops, schema_names):
+	known_tables = get_connection_tables(autogen_context, schema_names)
+
 	for schema in schema_names:
 		default = autogen_context.dialect.default_schema_name
 		if schema is None:
@@ -36,12 +31,11 @@ def compare_enums(autogen_context, upgrade_ops, schema_names):
 			metadatas = autogen_context.metadata
 		enums = get_declared_enums(metadatas, schema, default)
 		for table, values in enums.items():
-			fully_qualified_table_name = "%s.%s" % (schema, table) if schema else table
-			if is_table_present(fully_qualified_table_name, autogen_context.connection):
+			if (schema, table) in known_tables:
 				items = {
 					r[0]
 					for r in autogen_context.connection.execute(
-						"SELECT item_id FROM {}".format(fully_qualified_table_name)
+						f"SELECT {_ident('item_id')} FROM {_ident(schema)}.{_ident(table)}"
 					)
 				}
 				to_add = values - items
@@ -52,3 +46,41 @@ def compare_enums(autogen_context, upgrade_ops, schema_names):
 					upgrade_ops.ops.append(alembic_ops.EnumDeleteOp(schema, table, list(to_remove)))
 			else:
 				upgrade_ops.ops.append(alembic_ops.EnumInsertOp(schema, table, list(values)))
+
+
+def get_connection_tables(autogen_context, schema_names):
+	"""
+	Returns a sequence of (schema, tables) tuples that reflects the current state of the
+	database.
+
+	This is inspired by the code in alembic.autogenerate._autogen_for_tables
+	and alembic.autogenerate._compare_tables.
+	"""
+	inspector = autogen_context.inspector
+	default_schema_name = autogen_context.dialect.default_schema_name
+
+	conn_table_names = set()
+
+	version_table_schema = (
+		autogen_context.migration_context.version_table_schema
+	)
+	version_table = autogen_context.migration_context.version_table
+
+	for schema_name in schema_names:
+		if schema_name is None:
+			schema_name = default_schema_name
+		tables = set(inspector.get_table_names(schema=schema_name))
+		if schema_name == version_table_schema:
+			tables = tables.difference(
+				[autogen_context.migration_context.version_table]
+			)
+
+		conn_table_names.update(
+			(schema_name, tname)
+			for tname in tables
+			if autogen_context.run_name_filters(
+				tname, "table", {"schema_name": schema_name}
+			)
+		)
+
+	return conn_table_names