runner.go: Handle truncation of tokens for stop sequences

When a single token contains both text to be return and a stop
sequence, this causes an out of bounds error when we update the
cache to match our text. This is because we currently assume that
the removing the stop sequence will consume at least one token.

This also inverts the logic to deal with positive numbers, rather
than a value to be subtracted, which is easier to reason about.

Fixes #7153

llama/runner/stop_test.go

@@ -7,42 +7,54 @@ import (
 
 func TestTruncateStop(t *testing.T) {
 	tests := []struct {
-		name     string
-		pieces   []string
-		stop     string
-		expected []string
+		name          string
+		pieces        []string
+		stop          string
+		expected      []string
+		expectedTrunc bool
 	}{
 		{
-			name:     "Single word",
-			pieces:   []string{"hello", "world"},
-			stop:     "world",
-			expected: []string{"hello"},
+			name:          "Single word",
+			pieces:        []string{"hello", "world"},
+			stop:          "world",
+			expected:      []string{"hello"},
+			expectedTrunc: false,
 		},
 		{
-			name:     "Partial",
-			pieces:   []string{"hello", "wor"},
-			stop:     "or",
-			expected: []string{"hello", "w"},
+			name:          "Partial",
+			pieces:        []string{"hello", "wor"},
+			stop:          "or",
+			expected:      []string{"hello", "w"},
+			expectedTrunc: true,
 		},
 		{
-			name:     "Suffix",
-			pieces:   []string{"Hello", " there", "!"},
-			stop:     "!",
-			expected: []string{"Hello", " there"},
+			name:          "Suffix",
+			pieces:        []string{"Hello", " there", "!"},
+			stop:          "!",
+			expected:      []string{"Hello", " there"},
+			expectedTrunc: false,
 		},
 		{
-			name:     "Middle",
-			pieces:   []string{"hello", " wor"},
-			stop:     "llo w",
-			expected: []string{"he"},
+			name:          "Suffix partial",
+			pieces:        []string{"Hello", " the", "re!"},
+			stop:          "there!",
+			expected:      []string{"Hello", " "},
+			expectedTrunc: true,
+		},
+		{
+			name:          "Middle",
+			pieces:        []string{"hello", " wor"},
+			stop:          "llo w",
+			expected:      []string{"he"},
+			expectedTrunc: true,
 		},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := truncateStop(tt.pieces, tt.stop)
-			if !reflect.DeepEqual(result, tt.expected) {
-				t.Errorf("truncateStop(%v, %s): have %v; want %v", tt.pieces, tt.stop, result, tt.expected)
+			result, resultTrunc := truncateStop(tt.pieces, tt.stop)
+			if !reflect.DeepEqual(result, tt.expected) || resultTrunc != tt.expectedTrunc {
+				t.Errorf("truncateStop(%v, %s): have %v (%v); want %v (%v)", tt.pieces, tt.stop, result, resultTrunc, tt.expected, tt.expectedTrunc)
 			}
 		})
 	}