Measure bulk-onboarding write throughput: 10k-agent org = ~20-30k sequential kt=1 KEL appends

[bordumb]

Split out from #254's scale discussion (target: 10,000+ agents per org, ~1,000-engineer customer).

The ceiling

Caching fixes the read path; this is the write path. Onboarding 10,000 agents appends ~20–30k events to ONE org KEL, and KERI ordering serializes that chain: one git commit chain, one writer, strictly sequential seq numbers. At ~50ms/event that is ~25 minutes of pure sequential writes for day-one onboarding — possibly tolerable as a one-time batched import, but nobody has measured it.

What to do

1. Measure end-to-end append throughput for a synthetic 10k-agent onboarding against the Git-backed registry, with and without commit_batch (the batch path exists on RegistryBackend). Get a real events/sec number on launch-class hardware.
2. Decide whether batched import needs a dedicated bulk-onboarding flow (e.g. one anchoring ixn carrying many seals per batch, if the event algebra permits) or whether measured throughput is acceptable as-is.
3. Note the risk interaction: 1,000 engineers under a single kt=1 org KEL raises the operational stakes of the documented duplicity / concurrent-rotation accepted risk (docs/architecture/multi_device_accepted_risks.md) — same root scale, different failure mode. The threshold upgrade path in essays/design/multi_device.md becomes more load-bearing at this org size.

Acceptance

• A written events/sec number from a reproducible benchmark.
• A go/no-go on a dedicated bulk-import flow, recorded here.
• The duplicity-risk interaction acknowledged in the accepted-risks doc if the 10k-agent target is confirmed.

[bordumb]

Folding in two good additions from the now-closed duplicate #256:

1. Benchmark tiers: measure at N=100 / 1,000 / 10,000 agents, not just the top end.
2. Batching design candidate: beyond commit_batch, evaluate merging the scope seal into the anchoring ixn — one org-KEL event per agent instead of two — with a verify-first caveat (the verifier and read_agent_scope must accept seals on the anchoring event before any writer emits that shape).

[bordumb]

Prerequisite shipped in v0.1.3 (milestone 2 N+1 KEL elimination). Throughput measurement can now run against a released binary: auths 0.1.3.