Split from  bug 669727 .

Currently, if the user has the cursor in the middle of a string and makes an edit, triggering a new fetching of omnibox suggestions, we allow the user's text to split at the cursor.  In other words, we take the position of the cursor as interesting and try searching for both the text in the omnibox and the text in the omnibox as if there was a space character at the cursor.

This can result in unnecessary work if there is already a space character adjacent to the cursor.

That said, I *think* the caching done in url_index_private_data probably means these redundant calls take little to no time (because all the search terms are the same and so one pass should have lots of cache hits).