Benchmark harness scaffolding

[MaximilianRau04]

Description

docsgraph-eval needs a standalone harness to run benchmarks against docsgraph/docsgraph-server independently of the product repos, and report results in a comparable way over time.

Proposed Solution

Set up a runner that can load a benchmark suite, execute it against a configured target instance, and produce a structured, comparable results report.

Acceptance Criteria

• Harness can run a benchmark suite against a configured docsgraph/docsgraph-server instance
• Results are output in a structured, diffable format (e.g. JSON/CSV)
• Harness can be run locally and in CI
• README documents how to add a new benchmark suite