feat: ship API shell on fallback stub, resolve Vercel/PyTorch hosting, decision, pin deps, update docs

[hammadabbasiisi786-cloud]

• Verify API shell (Starlette app + Vercel functions) builds and runs
  end-to-end against FallbackSolver alone, no GROQ_API_KEY or checkpoint
  required; keep torch/groq imports lazy so serverless functions stay
  import-safe

• Confirm via tests/integration/test_solve_endpoint.py that the stub
  solver path returns correct shape/status for diff, integrate,
  partial, gradient, and tangent_line

• Resolve PyTorch-on-Vercel feasibility: neural inference (torch +
  model.pkl) exceeds Vercel's serverless deployment size limit and
  is not viable in prod; document decision in docs/HOSTING_DECISION.md
  (fallback-only in prod, Groq as opt-in upgrade, self-hosted neural
  deferred)

• Pin requirements.txt to exact versions for reproducible Vercel
  builds; add requirements-neural.txt for local/dev-only neural setup
  (was referenced in README but missing)

• Fix vercel.json build step that referenced an empty website/ dir
  and would have broken deployment

• Update README.md and docs/RUN_LOCALLY.md to reflect fallback-first,
  Groq-optional behavior and link the new hosting decision doc

[vercel]

@hammadabbasiisi786-cloud is attempting to deploy a commit to the chuzaairforce-7557's projects Team on Vercel.

A member of the Team first needs to authorize it.