The overhead of checking for Timeouts during a full-text query is rather high.

We should not verify this when accessing the already scored documents array: it's very unlikely to be useful and prevents inlining. We already perform the checks when actually scoring a new block of results, and also periodically when extracting entities.