fix(api): deterministic seed for collaborator notification test (flaky on main)

[raymondjacobson]

Fixes the flaky TestTrackCollaboratorNotificationsGenerated failing on main.

Root cause

The notification trigger's "new pending invite" branch fires only when created_at = updated_at (this is how it tells a fresh ETL insert from a reconcile re-write — the ETL writes them equal on insert). seedCollaborators didn't set those columns, so it inherited the seed baseRow defaults, which call time.Now() twice:

"created_at": time.Now(),
"updated_at": time.Now(),

After the timestamp column truncates to microseconds, those two calls are equal only if both landed in the same microsecond — a per-CI-run coin flip (the baseRow is a package var, initialized once). My local/isolated runs won the toss; CI lost it, so the trigger didn't fire and the test saw 0 notifications. (I'd earlier misattributed this to the unrelated TestSearch Elasticsearch panic — apologies.)

Fix

Pin a single now for both created_at and updated_at in the seed, so the invite branch always fires. Deterministic — passes 5× in a row locally, plus the full collaborator suite.

Note

The shared track_collaborators baseRow in database/seed.go still has the two-time.Now() default (latent, same as grants); not touched here to keep the fix scoped, since the only seeder that depends on the equality is now explicit.

🤖 Generated with Claude Code