[sehr-m]

Migrate the Autofix issue-data judge off the soon-to-be-deprecated /v1/llm/generate endpoint as part of consolidating all LLM prompts and model selection in Seer. Depends on the merged Seer one-shot handler in https://github.com/getsentry/seer/pull/8967.

Summary

• call the named autofix_issue_data_judge one-shot with the raw issue data
• remove prompt construction, issue-data filtering, model selection, and legacy response parsing from Sentry
• validate Seer's structured judge result before persisting the review
• preserve system viewer context and the existing task retry and stale-event behavior
• remove the Autofix caller from the S030 legacy API allowlist
• rely on the Seer handler's model_training feature attribution for proxy cost reporting

Validation

• .venv/bin/pytest -svv --create-db tests/sentry/tasks/seer/test_autofix.py -k TestAutofixIssueDataJudge (6 passed)
• .venv/bin/pytest -q tests/tools/test_flake8_plugin.py -k S030 (7 passed)
• SENTRY_MYPY_PRE_PUSH=1 .venv/bin/prek run -q mypy --files src/sentry/tasks/seer/autofix_issue_data.py --stage pre-push
• .venv/bin/prek run -q
• repository commit hooks
• pre-push mypy checks