feat: add persistent community navigation rail

[kalvinnchau]

Summary

• Replace the header community picker with a persistent left rail for Personal space, saved communities, and Add.
• Load NIP-11 icons through the same-origin broker without opening inactive relay sessions; validate image sources and fall back to saved artwork or initials.
• Preserve selection, join, navigation, and keyboard focus behavior.

Screenshots

Both web images use the same 1440×950 Chromium fixture; only the left 720 px are shown. The desktop image was supplied in the originating thread.

Before (web, a47ddaf)

After (web, rail)

Desktop reference

Verification

• Rebased head 4f2b99e: 2,474 Vitest tests; TypeScript, design checks/build; 84 Chromium/WebKit browser cases for affected layout, navigation, join, settings, scroll and unread-sidebar flows.
• New browser coverage: rail persistence, Personal/saved selection, Add focus restoration, and active/inactive NIP-11 icon loading without acquiring inactive sessions. Existing switcher journeys were converted; obsolete picker unit/gallery coverage removed and rail unit/icon tests added.

Deferred

• No community unread totals or badges: bounded observations cannot establish exact counts; product semantics and authorization for inactive session observation remain undecided.
• Leave/removal remains separate. Native desktop behavior and live relay icon fetching are not verified by the fixture screenshots.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

Reviewed 203c95c0148a0d078a803562bd2b9ae0ea840de2 against 222605777f8808617026ea33a0aad266bdaff4c6.

Changes requested: one P2 foreground-latency regression, detailed inline. Bound optional icon discovery so unavailable saved communities cannot monopolize the broker’s browser connections, and cover foreground progress while icon responses remain gated in Chromium and WebKit.

Selection/join ownership and icon URL/security review found no additional blockers. Hosted JavaScript and completed browser checks are green; the focused local browser/HTTP probe nevertheless reproduces starvation in both engines. CI exercises synthetic merge 487be15, not the isolated source head. Native desktop and live-relay behavior remain unverified. Unread totals and community removal remain outside scope.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

Re-review: no remaining blockers. Reviewed 14325bd1ed079400d21f019648a86e27d5368012 against base 222605777f8808617026ea33a0aad266bdaff4c6, concentrating on the fix since 203c95c.

The original P2 is resolved: two bounded icon workers leave foreground connection capacity, advance after failures, and stop scheduling/updating after effect cancellation. The new real-server regression requires a browser request to reach the server and complete while icon responses remain held, then verifies all eight discovery attempts arrive after release. I independently reviewed the worker lifecycle and integrated a second review of the test’s gate/order.

Hosted artifacts confirm the new test passed in Chromium (713ms) and WebKit (1,470ms); all current required CI checks are green. CI tested synthetic merge 8c7890f6ab236566f90376cb9c1ff5d7852eb094, incorporating this head into fe897a1. Broad suites were not duplicated locally. Native desktop/live-relay behavior remains unverified, and the synthetic foreground probe is not a full-app latency benchmark.

Withdrawing the prior changes request because its exit criteria are met. This is a review result, not an approval or merge.

Carl, an automated reviewer, commenting via Wes’s GitHub account. The sole P2 is resolved at 14325bd; bounded workers and passing gated Chromium/WebKit regressions meet the original exit criteria. See the verified follow-up review. No approval or merge.