Stop avoiding adding what the user typed to candidates

There does not seem to be any reason other than a historical
one to avoid doing this, but it takes processing power and
makes things more complicated.

This has a very limited impact on regression tests:
5 -> 3 [He,           the]
5 -> 3 [An,           an]
5 -> 3 [Where,        where]
5 -> 3 [This,         this]
7 -> 1 [wAtch,        watch]
6 -> 4 [oveNs,        oceans]
5 -> 1 [Ahere,        Where]
7 -> 1 [Hast,         Hast]
7 -> 5 [bjp,          bill]
5 -> 1 [What,         What]
5 -> 3 [Sound,        So und]
7 -> 3 [causalities,  casualties]
7 -> 3 [discontentment, discontent]
7 -> 3 [irregardless, regardless]

5 -> 1 : 2
5 -> 3 : 5
6 -> 4 : 1
7 -> 1 : 2
7 -> 3 : 3
7 -> 5 : 1

+1       4
-1       0
+2       0
-2       0
+3       8
-3       0
+4       1
-4       0
+5       1
-5       7
+6       0
-6       1
+7       0
-7       6

Change-Id: I6407cf922f27bbd3992df11d63690e71fc61111b

1 files changed, 1 insertions, 0 deletions

diff --git a/native/src/defines.h b/native/src/defines.h
index 48f92d86c..ce3f85acd 100644
--- a/native/src/defines.h
+++ b/native/src/defines.h
@@ -172,6 +172,7 @@ static void prof_out(void) {
 #define EQUIVALENT_CHAR_WITHOUT_DISTANCE_INFO -2
 #define PROXIMITY_CHAR_WITHOUT_DISTANCE_INFO -3
 #define NOT_A_INDEX -1
+#define NOT_A_FREQUENCY -1
 
 #define KEYCODE_SPACE ' '