Capture finish time and failure reason on every test run

[soulgalore]

The /admin "Recent failures" table reads straight from Bull's
retained-failures list, which only sees jobs the testrunner threw — not
test-level failures (sitespeed.io exits non-zero) or queue-down failures at
submit time. The DB knows about all three modes via status='failed' but didn't
store why a test failed or when it reached a terminal state, so the operator
UI shows a health pill saying "1 failed in 24h" with an empty table
underneath. Accurate but unactionable.

This change makes the database authoritative for the failure picture. Two new
columns are populated from every path that flips a row to 'failed' (Bull
global:failed, queue-down submit, testrunner exit on the result queue), and
finished_date is also stamped on completed runs so end-to-end and run-only
durations become measurable — run_date today is the browsertime start
timestamp, not finish. The follow-up that switches the admin table to read
from the DB lands separately.

The major bump is because existing deployments must run the included migration
before rolling forward; it is metadata-only on Postgres 11+ so it runs in
milliseconds, and ships with a sibling shell helper that wraps the
docker-compose psql invocation.

Co-authored-by: Claude noreply@anthropic.com