Refresh test schema to match current platform schema

[rowanseymour]

The hand written testdb.sql used by the test suite had drifted from the platform schema it's meant to mirror. Since it's only a subset of that schema, drift is easy to miss — but a column that's misnamed, wrongly typed, or has the wrong nullability means the tests can pass against a shape that doesn't exist in production.

Refreshed it against the current schema. The rule is now stated at the top of the file: it stays a minimal subset (only the tables and columns archiver reads from or deletes from), but any column that is present must match its real definition exactly.

Changes

Renamed or removed upstream:

• auth_user no longer exists — replaced with users_user (username → email)
• msgs_msg.quick_replies varchar(64)[] → quickreplies jsonb
• added flows_flowstartcount, which DeleteFlowStarts deletes from but which was missing entirely
• dropped the DROP TABLE lines for channels_channellog (no longer exists) and ivr_call (never created here)

Nullability, two of which were inverted:

• flows_flowrun.session_uuid is NOT NULL, flows_flowrun.results is nullable
• msgs_msg.next_attempt, msgs_broadcast.is_active, contacts_contact.created_by_id and modified_by_id are all nullable

Types:

• uuid rather than varchar(36) on channels_channel, flows_flow, contacts_contactgroup, msgs_label (contacts_contact.uuid is genuinely still a varchar)
• bigint/bigserial on archives_archive.id and rollup_id, flows_flowrun.id, msgs_label.id, msgs_msg_labels.*, msgs_broadcast.schedule_id, msgs_broadcastmsgcount.count and the m2m table ids
• msgs_msg.attachments is varchar(2048)[]
• name lengths on orgs_org, channels_channel, flows_flow, contacts_contactgroup

Also added the non-null columns needed to keep the fixture coherent (msgs_label.org_id, flows_flowstart.flow_id/uuid/start_type/status/modified_on, msgs_broadcast.base_language, msgs_broadcastmsgcount.is_squashed) plus the foreign keys on archives_archive.org_id and rollup_id.

Coverage for orphaned flow start deletion

The reason the missing flows_flowstartcount table went unnoticed is that its delete path never ran: the fixture's only flow start had a run pointing at it, so the NOT EXISTS guard in DeleteFlowStarts never selected anything and the DELETE from flows_flowstartcount statement never executed.

Added flow start #2 for org 3 — old enough to be past the retention threshold, with no runs pointing at it — along with its contact, group and count rows, and an org 3 contact group for it to reference. TestArchiveOrgRuns now asserts all four rows are deleted while start #1, which still has a run, survives.

Test plan

• go test -p=1 ./... passes.
• Verified the schema refresh by diffing information_schema.columns for the loaded test schema against the real one — every column present matches on name, type, length and nullability.
• Verified the new assertions are meaningful by flipping the flows_flowstartcount count expectation to 1 and confirming TestArchiveOrgRuns fails, so the delete really is executing rather than the row never having existed.
• Fixture data is otherwise unchanged apart from values for the newly required columns, so the archive content and hash assertions are untouched.

[claude]

Review Summary

Solid, well-documented change — this is a schema-only refresh of the testdb.sql fixture to match the real temba schema, with clear rationale in the description. I verified the main risks and found no correctness issues:

• Renamed/retyped columns don't break queries. quick_replies → quickreplies is safe: no Go code references either name (the message archive query at archives/messages.go reads only attachments, which stays varchar(...)[] and works with unnest). The auth_user → users_user rename is reflected in the msgs_msg.created_by_id FK.
• New NOT NULL columns have no conflicting fixture data. msgs_optin, msgs_broadcastmsgcount, and flows_flowstartcount aren't seeded, and the flows_flowstart / msgs_broadcast / msgs_label inserts were updated for their new required columns.
• FK & create ordering are correct — flows_flowstartcount is created after flows_flowstart and dropped via CASCADE.
• DeleteFlowStarts now has a table to delete from (flows_flowstartcount), which was previously missing entirely.

One gap (also acknowledged in the PR notes): the newly-added flows_flowstartcount table and the DeleteFlowStarts delete path are still never exercised by the tests, since the fixture's only flow start has a run attached. Adding an orphaned start would both close that coverage gap and protect the new table from silent future drift — flagged inline.

No blocking issues.