Infrastructure: commanded-vs-baseline benchmark harness for SOTA bots

[minsing-jin]

Goal

Make SOTA bot experiments comparable through a commanded-vs-baseline benchmark harness.

Tasks

• Define match matrix: bot, opponent, race, map, command script, baseline script, telemetry output, replay output.
• Add generated run plans for each benchmark case.
• Support command scripts for aggressive, defensive, greedy, harass, contain, and no-all-in styles.
• Normalize replay-derived metrics into report inputs.
• Produce comparison reports with fulfillment delta, adherence delta, win/loss, crash/desync markers, and command override counts.

Acceptance Criteria

• Benchmark harness can emit plans for each candidate bot without live execution.
• Harness separates commandable bots from benchmark-only opponents.
• Every generated case specifies queue path, telemetry path, replay/log path, and expected verifier metrics.
• Docs explain how to run the same case under BWAPI/BASIL/SSCAIT-style local runners.