Drops Python 2 support

REP-1030

In addition to some python 2 => 3 fixes, this change bumps the scikit-learn
version to latest. The previously pinned version of scikit-learn failed trying
to compile all necessary C modules under python 3.7+ due to included header files
that weren't compatible with C the API implemented in python 3.7+.

Simultaneously, with the restrictive compatibility supported by scikit-learn,
it seemed prudent to drop python 2 support altogether. Otherwise, we'd be stuck
with python 3.4 as the newest possible version we could support.

With this change, tests are currently passing under 3.9.2.

Lastly, imports the original training data. At some point, a new version
of the training data was committed to the repo but no classifier was
trained from it. Using a classifier trained from this new data resulted
in most of the tests failing.

setup.py

@@ -19,12 +19,12 @@ class InstallCommand(install):
         if self.no_ml:
             dist = self.distribution
             dist.packages=find_packages(exclude=[
-                'tests',
-                'tests.*',
-                'talon.signature',
-                'talon.signature.*',
+                "tests",
+                "tests.*",
+                "talon.signature",
+                "talon.signature.*",
             ])
-            for not_required in ['numpy', 'scipy', 'scikit-learn==0.16.1']:
+            for not_required in ["numpy", "scipy", "scikit-learn==0.24.1"]:
                 dist.install_requires.remove(not_required)
 
 
@@ -48,12 +48,13 @@ setup(name='talon',
           "regex>=1",
           "numpy",
           "scipy",
-          "scikit-learn==0.16.1", # pickled versions of classifier, else rebuild
-          'chardet>=1.0.1',
-          'cchardet>=0.3.5',
-          'cssselect',
-          'six>=1.10.0',
-          'html5lib'
+          "scikit-learn==0.24.1", # pickled versions of classifier, else rebuild
+          "chardet>=1.0.1",
+          "cchardet>=0.3.5",
+          "cssselect",
+          "six>=1.10.0",
+          "html5lib",
+          "joblib",
           ],
       tests_require=[
           "mock",