When searching for issues in the issue picker, if someone enters a prefix that matches a lot (many thousands) of issue keys, this gets expanded by lucene into a query with thousands of terms. As this search is unsorted, it then tries to score it, creating a scorer per term. These scorers take up a lot memory (we saw 80MB for 40K issues) and are completely unnecessary as we will only ever match one issue key in the issue key field.

We need to implement a ConstantScorePrefixQuery, similar in design to Lucene's ConstantScoreRangeQuery.