perf(pulse-core): benchmark suite with 20% regression gating (#924)

[lauferuqquyah]

Adds a performance benchmark suite for pulse-core's hot path, with CI
regression gating at 20% against a committed baseline.

What's included

• packages/pulse-core/bench/ - dependency-free harness (node:perf_hooks),
  the four required benchmark cases, the regression gate, and the entry point.
• Cases: raw event -> NormalizedEvent throughput, decode-with-spec throughput,
  watcher fan-out at 1/100/1000, and cursor write cost per adapter.
• Runs against the recorded CAP-67 fixture corpus, so it measures pulse-core,
  not the network.
• Results stored as JSON, compared against a committed baseline in CI; a
  regression worse than 20% fails the build.
• New benchmarks job in ci.yml, and a Performance section in
  docs/ARCHITECTURE.md documenting the suite and the baseline-update process.
• Removes the two prior ad-hoc bench scripts (throughput.ts,
  soroban-throughput.ts), which measured throughput but did not gate.

Baseline note

The committed baseline.json was generated on a developer workstation and is
noisy (high run-to-run variance). It should be regenerated on stable hardware
before the gate is relied on - the first CI run on a dedicated runner is a good
source. This is the reseed process documented in ARCHITECTURE.md.

Closes #924

[vercel]

@lauferuqquyah is attempting to deploy a commit to the determined's projects Team on Vercel.

A member of the Team first needs to authorize it.

[drips-wave]

@lauferuqquyah Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits

[lauferuqquyah]

hello please merge