Reduce key dictionary memory footprint

Instead of indexing `Key` objects that have multiple names by each
individual name, index by the set of names. This reduces the size of the
default key dictionary by a factor of between 2 and 3, and as result
also reduces reallocations/defragmentation.
Instead of instantiating all module/extension keys by default, append
them to the `maybe_key`-generator list.

kmk/handlers/sequences.py

@@ -35,7 +35,7 @@ def simple_key_sequence(seq):
         meta=KeySequenceMeta(seq),
         on_press=sequence_press_handler,
         on_release=passthrough,
-    )
+    )[0]
 
 
 def send_string(message):
@@ -124,7 +124,7 @@ def unicode_codepoint_sequence(codepoints):
                 simple_key_sequence(_winc_unicode_sequence(kc_macros, keyboard)), True
             )
 
-    return make_key(on_press=_unicode_sequence)
+    return make_key(on_press=_unicode_sequence)[0]
 
 
 def _ralt_unicode_sequence(kc_macros, keyboard):