Adds support to serialize dict as JSON

There are likely some vendor-specific enhancements (such as `JSONB` in PostgreSQL) but the current method of type guessing doesn't know anything about the underlying database.

dataset/types.py

@@ -1,7 +1,7 @@
 from datetime import datetime, date
 
 from sqlalchemy import Integer, UnicodeText, Float, BigInteger
-from sqlalchemy import Boolean, Date, DateTime, Unicode
+from sqlalchemy import Boolean, Date, DateTime, Unicode, JSON
 from sqlalchemy.types import TypeEngine
 
 
@@ -15,6 +15,7 @@ class Types(object):
     boolean = Boolean
     date = Date
     datetime = DateTime
+    json = JSON
 
     def guess(self, sample):
         """Given a single sample, guess the column type for the field.
@@ -34,4 +35,6 @@ class Types(object):
             return self.datetime
         elif isinstance(sample, date):
             return self.date
+        elif isinstance(sample, dict):
+            return self.json
         return self.text

test/test_dataset.py

@@ -192,6 +192,20 @@ class TableTestCase(unittest.TestCase):
             )
         assert len(self.tbl) == len(TEST_DATA) + 1, len(self.tbl)
 
+    def test_insert_json(self):
+        last_id = self.tbl.insert({
+            'date': datetime(2011, 1, 2),
+            'temperature': -10,
+            'place': 'Berlin',
+            'info': {
+                'currency': 'EUR',
+                'language': 'German',
+                'population': 3292365
+            }
+        })
+        assert len(self.tbl) == len(TEST_DATA) + 1, len(self.tbl)
+        assert self.tbl.find_one(id=last_id)['place'] == 'Berlin'
+
     def test_upsert(self):
         self.tbl.upsert({
             'date': datetime(2011, 1, 2),