Fixes the issue in #4373

Problem definition

• Many languages do sprints, workshops, etc where multiple people come at the same time and do the same thing. Here, sentence validation.
• The system gets 100 not-yet-validated sentences from the database with specific conditions (previous user votes and variants included - but these sprints are already done for those language variants).
• So, if 20 people ask for unvalidated sentences, they get exactly the same 100 sentences in exactly the same order. They are kept in the client (browser) without knowing changes on the server.
• When they validate, each sentence will have (mostly) same 20 votes.
• And this will result in loss of many valuable volunteer effort. 20 people would validate only 100 sentences.

Solution provided

• Select 1000 sentences (somewhat deterministic again)
• Select 100 sentences among them randomly.

Of course there can be some collisions for more than needed votes, but this change drops the possibility roughly to 10%. We might say 20 people would validate 900 sentences in that time, a 900% increase.

Edit: Tested (and fixed) on local dev env