feat: Retry individual runs on platform inference errors

[thomasvangurp]

Summary

• Builds on feat: Detect platform-side inference errors #332 by @statxc (platform inference error detection)
• Instead of restarting the entire evaluation when inference errors exceed the threshold, retries only the specific failed run
• Addresses feedback from @ibraheem-abe in feat: Detect platform-side inference errors #332: "We want to change this so only that specific test is retried instead of the entire thing"

What changed

File	Change
validator/main.py	Retry loop in _run_evaluation_run(): on inference error threshold, reset counter and retry just this run (up to 2 retries). Only marks as platform error (3050) after exhausting retries.
inference_gateway/error_hash_map.py	Added reset_inference_errors() method to clear error count before retry
inference_gateway/main.py	Added POST /api/reset-inference-errors endpoint
tests/test_inference_error_tracking.py	Added tests for reset behavior

How it works

Agent runs problem X → hits 5 inference errors → threshold exceeded
  → Validator resets error counter
  → Validator retries ONLY problem X (not the whole evaluation)
  → If retry also fails → one more retry
  → If all 3 attempts fail → mark as platform error (3050)
Meanwhile, problems Y and Z continue running normally

Test plan

• Verify ErrorHashMap.reset_inference_errors() clears the count
• Verify reset doesn't affect other runs
• Verify retry loop retries on threshold exceeded
• Verify platform error raised after max retries exhausted
• Run python3 -m pytest tests/test_inference_error_tracking.py -v

🤖 Generated with Claude Code

[thomasvangurp]

@ibraheem-abe This implements the change you requested in #332 — retrying only the specific run instead of restarting the entire evaluation. Would appreciate your review when you get a chance. Thanks!