downstream data - 300 sample test

feat: add rich console output to downstream eval, populate README with extrinsic results

docs: restructure README with intrinsic/extrinsic evaluation framing

Merge pull request #24 from Sudhendra/downstream-eval-framework

fix: allow completed downstream resumes without tinker key

fix: validate tinker backend before dataset load

fix: remove stale yaml type ignores

fix: address downstream eval CI issues

fix: clarify downstream summary counts

fix: support autoformat hook in worktrees

feat: add downstream task evaluation pipeline

docs: add Nanbeige 3B equivalence results (1.3% pass rate, 3-gate breakdown)

docs: add equivalence eval 3-gate scoring diagram to README

docs: add per-model avg gate scores to Qwen3-8B breakdown

feat: 3-gate evaluation system, Qwen3-8B equiv results (2.0% pass rate)