Reproducible bench harness for collection performance

[priethor]

Goal

Rows endpoint performance is hard to compare between branches today. Add a WP-CLI benchmark that seeds the same dataset and reports numbers we can diff.

Scope

• wp cortext perf-seed creates a representative dataset. Sizes are controlled with flags for collections, rows, fields, relations, and rollups.
• wp cortext perf-bench runs the benchmark and emits JSON with p50/p95 latency, SQL query counts, and memory.
• Cover these scenarios: rows endpoint page 1 and page 50, sort by meta_value, edit a relation with many targets, and bulk migrate at the 1000-row cap.
• CI job runs the benchmark on a representative dataset and fails when any scenario exceeds its budget. Budgets live in a JSON file next to the bench so they're reviewable.
• Put the command in includes/CLI/PerfBench.php.

Acceptance criteria

• Seeding a clean install finishes in under 2 minutes.
• Benchmark output keeps the same shape across runs, so diffs stay readable.
• CI fails the build when a scenario exceeds its budget.