fix(desktop): stop hiding repos re-announced after a deletion

[daxdax89]

Fixes #3760.

Problem

isDeletedByA hid a project card whenever any same-author kind:5 tombstone carried its 30617:<owner>:<dtag> coordinate, with no timestamp comparison:

event.pubkey.toLowerCase() === project.owner.toLowerCase() &&
event.tags.some((tag) => tag[0] === "a" && tag[1] === coordinate)

Repo announcements are addressable, so the coordinate outlives any single event at it. Deleting a repo and announcing the same dtag again is a legitimate recovery path, and NIP-09 scopes an a-tag deletion to versions "up to the created_at timestamp of the deletion request event" — a tombstone must not hide an announcement published after it.

The relay already gets this right, which is what made the failure confusing to diagnose: the newer announcement is stored, served over REQ, and clone/push works, while Desktop's Repositories list never renders the card again for anyone. Once an owner had ever deleted a project (including via Desktop's own Delete project button), no re-announcement at that coordinate could come back.

Fix

Bound the match to event.created_at >= project.createdAt, keeping the existing author check. The comparison is inclusive, matching NIP-09's "up to the created_at timestamp".

The predicate and projectCoordinate move from hooks.ts into lib/projectDeletions.ts. Two reasons: the predicate was unexported and therefore untestable, and hooks.ts was at 997 of the 1000 lines desktop/scripts/check-file-sizes.mjs allows, so the fix could not have been added in place. hooks.ts is now 986 lines and both call sites (fetchProjects, fetchProject) are unchanged apart from the import.

I kept the isDeletedByA name so it still matches the issue.

Tests

lib/projectDeletions.test.mjs — 8 cases covering the regression (tombstone older than the announcement no longer hides it), a newer tombstone still hiding it, and the inclusive same-timestamp boundary. The author check, case-insensitive pubkey match, and coordinate matching already worked and are pinned so a later refactor cannot quietly drop them.

Verified the suite catches the bug: with the created_at comparison removed, exactly one test fails — "a tombstone older than the announcement does not hide it".

Checks

just desktop-check, just desktop-typecheck, just desktop-test (3827 pass), and just desktop-build all pass locally.

Out of scope

The issue's second half is deliberately not addressed here, since it needs the managed-agent ownership relationship on the client and is a larger change:

• isDeletedByA only honors deletions whose pubkey equals the announcement author, so a relay-accepted owner-of-agent deletion (is_agent_owner in validate_standard_deletion_event) still leaves the card rendered.
• canDelete in ProjectsView.tsx gates on strict key equality via isProjectOwnedByCurrentUser, so a user cannot delete their own managed agent's repos from Desktop even though the relay would authorize it.

Happy to follow up on those separately if useful.

Related PRs

Searched open PRs and issues for duplicates; found none touching isDeletedByA or repo-deletion filtering.

[Chessing234]

good catch on re-announced repos. can you add a test where created_at equals the deletion time so the boundary is pinned?