In this test, it's impossible that start < 0 so the test is useless.
I'm not sure what the cursor test was for, but it's very old code, and
it seems the assumption was either misled or doesn't hold any more:
testing for the absolute cursor position against the length of the
word against the cursor makes no sense.

The net result of this was that when the cursor index got large
enough, resuming suggestion would not work any more.

Bug: 7586467
Change-Id: I3462082374fe9579bec7698f4d424de6ff5f2ded