Restore uniform random generator challenge sampling

[dylanuys]

Summary

• Default challenge targeting is a uniform random draw of --neuron.sample-size unique generators (image/video assigned at random).
• Qualified / onboarding / probe slot allocation remains available via --neuron.challenge-allocation buckets.
• Qualification still gates pay only; it no longer decides who gets asked each round.

Test plan

• pytest tests/test_challenge_allocation.py
• Confirm a validator log line Challenge allocation: random 50/<pool> generators
• Confirm previously starved UIDs appear in challenge rounds over a few cycles
• Optional: --neuron.challenge-allocation buckets still fills 36/8/6 as before

Made with Cursor