chore(ci): extract E2E fixture seeding to a shared script instead of scraping deploy.yml #203

Follow-up to #198 (merged as d36d1ecc). Raised in review, deliberately not blocking that merge, filed so it is not lost.

The problem

scripts/e2e-local.sh gets its fixture-seeding SQL by parsing .github/workflows/deploy.yml with grep/sed/awk and piping the result into bash and sqlite3:

WF=.github/workflows/deploy.yml
grep -E "sqlite3 test-fixtures/e2e-fixture.db \"UPDATE observers" "$WF" \
  | sed -E 's/^ *run: *//; s#test-fixtures/e2e-fixture.db#'"$DB"'#' | bash
awk '/sqlite3 test-fixtures\/e2e-fixture.db <<.SQL.$/{f=1;next} /^ *SQL$/{f=0} f' "$WF" | sqlite3 "$DB"

This couples local E2E to the formatting of the workflow rather than its behaviour. Any of these breaks it:

• re-indenting the run: block
• renaming the SQL heredoc delimiter
• changing the quoting on the UPDATE observers line
• splitting that step in two

The failure mode is silent. grep matches nothing, the pipeline into bash succeeds with empty input, set -euo pipefail does not catch an empty-but-successful pipeline, and the script proceeds with an under-seeded fixture. Local tests then fail for reasons that have nothing to do with the code under test — which is the same class of confusion #199/#201 existed to remove.

Why it was merged anyway

The blast radius is one developer's local run. Nothing in CI invokes scripts/e2e-local.sh, nothing on prod depends on it, and it cannot produce a bad release. Worst case is wasted debugging time. Holding a devcontainer that unblocks macOS development on a maintainability concern was poor value.

The fix, and it is a pattern already in the repo

#199 solved exactly this shape for the server start: duplicated inline logic was pulled into scripts/e2e-fresh-server.sh and deploy.yml was changed to call it. One definition, two callers.

Do the same for seeding:

1. Extract the seeding into scripts/e2e-seed-fixture.sh "$DB" — the UPDATE observers IATA repair and the #1486 grouped-packet row, i.e. the Seed observer IATA codes and Seed grouped-packet row for #1486 collapse test steps [repo: .github/workflows/deploy.yml:321,325].
2. Change those workflow steps to call it.
3. Change e2e-local.sh to call it, deleting the grep/sed/awk block.

Result: no YAML parsing, no silent empty-seed, and seeding changes land in one place for both CI and local.

Worth deciding while in there

Whether Migrate fixture DB to current schema (#1287) [repo: deploy.yml:312] should move into the same script. e2e-local.sh already runs its own corescope-migrate -baseline-stamp-migrate, so that logic is duplicated too — but the CI and local invocations differ slightly, so this is a judgment call rather than an obvious lift.

Done when

scripts/e2e-local.sh contains no reference to deploy.yml, both CI and the local runner seed the fixture through one shared script, and a formatting change to the workflow cannot silently under-seed a local fixture.

Refs #198, #200, #199, #201