perf: skip 3s DWN discovery for returning users and parallelize registration

[LiranCohen]

Summary

Rebased on current main and re-applied the still-valid startup improvements in the current Effect-based code.

Changes

• Skip local-DWN discovery during provider initialization when a same-tab vault password is already cached.
• Keep the current discovery helper behavior that resolves early when an endpoint appears.
• Parallelize DWN registration while preserving current network policy retries and injected registration-token storage.
• Prepare provider-auth tokens once per endpoint, then register all agent/identity DIDs concurrently.
• Add provider tests for cached-session discovery skip and fresh-session discovery.

Verification

• pnpm vitest run src/enbox/__tests__/provider.test.tsx src/enbox/__tests__/registration.test.ts
• pnpm lint (passes with existing Testing Library warnings)
• pnpm build
• pnpm test

[github-actions]

Coverage Report

Status	Category	Percentage	Covered / Total
🔵	Lines	58.42% (🎯 25%)	2260 / 3868
🔵	Statements	56.33% (🎯 25%)	2387 / 4237
🔵	Functions	58.09% (🎯 30%)	804 / 1384
🔵	Branches	53.55% (🎯 29%)	1364 / 2547

File Coverage

File	Stmts	Branches	Functions	Lines	Uncovered Lines
Changed Files
src/enbox/provider.tsx	67.27%	54%	72.41%	70.06%	90, 138, 153-166, 191, 196, 200, 206, 211, 221, 244-246, 255-279, 291, 309-311, 320-335, 344, 349-350
src/enbox/registration.ts	64.44%	51.61%	65.3%	65.9%	59, 64, 69-134, 158, 168-172, 209, 242, 252, 319-321, 344-345

Generated in workflow #350 for commit 299b66a by the Vitest Coverage Report Action

[LiranCohen]

Triage (2026-07-07): keeping this open — it's the only survivor of the stale-PR sweep. Branch was rebuilt on the Effect-based code on 2026-06-23, is currently MERGEABLE/CLEAN against main, and CI is green on the refreshed head. Premise (skip local-DWN discovery for returning users, parallelize registration) is startup-perf work orthogonal to the connect flow changes. Needs a review pass before merge — it hasn't had one since the rebuild.