Record run provenance (seed, k, wall-clock, hardware) in results summary

[bamdadd]

Context

Every rate is stochastic (k seeded repeats, bootstrap CIs), and CONTRIBUTING requires results to state seeds, hardware, and wall-clock. The written summary should carry that provenance so a results file is self-describing and reruns are comparable — some of this may already be emitted, so this is a gap-fill, not a rewrite.

Acceptance criteria

• The results summary JSON records, if not already present: base seed, k, bootstrap iters, wall-clock, and a coarse hardware/runtime tag (e.g. platform string).
• report output surfaces this provenance so a reader sees the run conditions next to the rates.
• A documented one-command repro path reproduces the committed stub numbers deterministically (fixed seed).
• Test asserts the provenance fields are present and that two stub runs at the same seed produce identical rates.