Run the dashboard's per-environment tests as a matrix job (26 min -> ~7)

[gaurav]

The dashboard workflow runs the validation suite against each environment in sequence, and that loop is essentially the entire runtime. From the most recent run (26 minutes end to end):

25m 48s   Run the test suite against every target
    30s   everything else — checkout, report generation, website build, deploy

Six targets at roughly four minutes each, one after another, on one runner. Splitting them into a matrix job — one target per runner, each uploading its raw/$t.jsonl as an artifact, with a single dependent job that downloads them all, generates the report and deploys — would bring the wall clock down to about the slowest single target plus overhead: roughly 6-8 minutes instead of 26.

dashboard.yaml already carries a ponytail: note anticipating this ("sequential loop; split into a matrix job with artifact merging if total runtime ever approaches the 6h job limit"). The 6h limit is not the reason to do it — the feedback loop is. A four-hour-old dashboard is fine; waiting 26 minutes to see whether a change to the report generator worked is not.

Points to get right:

• The matrix must be built from targets.ini, not hardcoded: read_targets() is the single source of truth as of Replace the website with a daily cross-environment test dashboard built around promotion drift #120. A fromJSON matrix fed by a small setup job that prints the target list would keep that property.
• continue-on-error per target, since a nonzero pytest exit is normal here — the report is the artifact.
• Only the final job should touch gh-pages; it force-pushes as a single commit.
• Keep timeout 45m per target, which caps a hung environment.

Deliberately not done in #120: it restructures the job graph and the deploy step, and it cannot be validated without a full run, so it wants its own review.