The release load run is one command a person can take on a copy: both passes, both result files with a verdict, production refused #9596

Replaces #9565. A crew sandbox can't reach a copy holding production's data, so the crew makes the person's run a single command and leaves the measuring to them.

Done looks like:

• [x] bun scripts/release-load.ts <version> --url <copy> --repo <owner/name> runs the pass through the edge and the pass that bypasses it, and writes bench/results/<version>.json and bench/results/<version>.bypass.json.
• [x] Each file records rate, refused, p50, p95, warm-up and "verdict": "target met" or how far it missed (target: 500/s, 0 refused, p95 under 100 ms).
• [x] A --url that resolves to production (the forge's own public host) is refused before any request goes out, with a test.
• [x] Tests against a local fake server cover both files and both verdicts.
• [x] docs/features/load-test.md says, step by step, how a person stands up the copy behind the edge and runs it, and where to post the files (#9514).

Files: scripts/release-load.ts, bench/load.ts, docs/features/load-test.md, test/release-load*.test.ts. Risks: never aim load at production; the crew writes no result file it didn't measure.

Reviewed pull #9598 at its head: ticked bun scripts/release-load.ts <version> --url <copy> --repo <owner/name> runs the pass through the edge and the pass that bypasses it, and writes bench/results/<version>.json and bench/results/<version>.bypass.json.; Each file records rate, refused, p50, p95, warm-up and "verdict": "target met" or how far it missed (target: 500/s, 0 refused, p95 under 100 ms).; A --url that resolves to production (the forge's own public host) is refused before any request goes out, with a test.; Tests against a local fake server cover both files and both verdicts.; docs/features/load-test.md says, step by step, how a person stands up the copy behind the edge and runs it, and where to post the files (#9514)..