perf(pipeline): Add sustained benchmark validation

[rahulbsw]

Summary

• add partition-ordered processing and exact delivery accounting for trustworthy
  throughput validation
• replace startup-inclusive measurements with separate ingress, output, and
  metrics-validator jobs plus sustained repetitions
• refresh the Criterion and on-demand throughput workflows with structured
  evidence artifacts
• consolidate the documentation and rebuild GitHub Pages as a curated
  open-source product site
• add branded site assets, link validation, benchmark methodology, and a
  demo-recording runbook

Why

The earlier throughput path mixed environment startup, data ingress,
processing, and validation. This change makes the environment warm before
timing, validates exact input/output counts, and records reproducible evidence
without presenting a universal throughput claim.

Validation

• cargo test --all: 474 passed, 0 failed, 30 ignored
• cargo clippy --all-targets --offline -- -D warnings: passed
• cargo fmt --all -- --check: passed
• benchmark helper unit tests: 6 passed
• benchmark shell syntax and JSON schema validation: passed
• Podman Compose configuration: passed; Kafka remains bound to 127.0.0.1
  and the benchmark network is internal
• local Jekyll build and generated-site internal-link validation: passed
• sustained local benchmark: three 120-second repetitions with exact
  record-count validation and zero processing errors; intentionally not
  promoted as a public headline result because the run was ingress-limited

Follow-up

Run the same sustained harness on a cost-capped AWS ECS-on-EC2 environment
provisioned as code, with no service exposed to the public internet.