Validate benchmark JSON with a versioned schema

[DaBestCode]

Make recorded benchmark results machine-checkable as the benchmark suite grows.

Acceptance criteria:

• Add a JSON Schema for schema_version: 1 results emitted by scripts/benchmark_transformers.py.
• Validate the checked-in Qwen3 microbenchmark during tests.
• Document how schema changes should be versioned.
• Reject missing device, model revision, protocol, summary, or claim-scope fields.
• Do not add a runtime dependency to the randkv package.

[faizan-7890]

Say: “I’d like to work on this. Can this be assigned to me?”

[faizan-7890]

Add a versioned JSON Schema for schema_version: 1 benchmark output from scripts/benchmark_transformers.py.
Add tests to validate the checked-in Qwen3 microbenchmark against this schema.
Document schema versioning/update rules.
Ensure validation rejects missing required fields (device, model_revision, protocol, summary, claim_scope).
Keep it dependency-free (no runtime dependency on randkv)

[DaBestCode]

Thanks for volunteering, @faizan-7890 — I’ve assigned this to you.

A few implementation details to remove ambiguity:

• Put the schema at schemas/benchmark-v1.schema.json and target JSON Schema Draft 2020-12.
• The current result has no literal protocol object. For schema v1, treat the protocol as the existing required fields: kind, prompt_tokens, generated_tokens, warmup_tokens, trials, budget, buffer_size, and seed. Please do not change the producer format solely to add a protocol key.
• Also require schema_version, claim_scope, model, model_revision, device, runs, and summary.
• jsonschema is fine as a test-only dependency under the test extra; the base randkv install must remain dependency-free.
• Add one positive test for the checked-in Qwen3 result and parameterized negative cases that remove each required field.
• Document compatibility rules in results/README.md: additive optional fields stay in v1; removing/renaming fields or changing their meaning requires a new schema version.

Please keep the PR focused on the schema, tests, test dependency, and documentation. Before opening it, run:

ruff check .
ruff format --check .
mypy src
pytest -q

Feel free to open a draft PR early if you want feedback on the schema structure.