Avoid lagging for edge cases while suggesting

[dimztimz]

Hunspell uses 3 techniques.

1. Add time limits in the slow algorithms (O(n^2) or more). Hunspell uses this in only 3 of all 12 algorithms. It does the time check only after it checks 100 candidates. If the time limit was exceeded, it stops, otherwise it goes on for another 100 attempts and checks the time limit again and so on.
2. Firstly, check the candidates of all 12 algorithms only as simple words, and only if there are no suggestions, rerun the same algorithms checking the candidates as compounds. This shuffling speeds up the case when there are suggestions, but it does not speed up the case when there are not any suggestions.
3. Limit the number of suggestions. Hunspell has a hard limit to 15 suggestions. From my experience this condition is very rarely hit, so it is the least effective. It is probably for some pathological cases.

[dimztimz]

I think this is almost solved with commit 8558564, anything more can be done in a future version.

[dimztimz]

Major improvements with cb3e4c5.