feat(ci): run the fixture release flow nightly

[danceratopz]

🗒️ Description

This makes the cron-trigger idea from the ethereum#3100 review thread concrete: instead of adding a parallel nightly workflow, the existing release_fixtures.yaml flow runs on a nightly schedule and stops after combine, so no tag or draft release is created. The nightly thereby becomes a daily rehearsal of the actual release pipeline: generate_build_matrix.py, build-fixtures, the index merge and the release tarball all get exercised every night, and the resulting fixtures_nightly.tar.gz artifact is release-shaped (merged .meta/index.json included), so it can be consumed directly.

Scheduled runs fill a new nightly feature in feature.yaml (--until=Amsterdam --generate-all-formats, slow tests included) and skip themselves when there are no new commits since the last successful scheduled run: the baseline query filters on event=schedule so manual releases never advance it, and a failed nightly keeps re-running until it goes green. fail-fast is disabled on schedule so every fork range reports, run-name distinguishes "Nightly Fill" runs from manual releases in the run history, and artifact retention (5 days for the combined tarball, 1 day for the per-range splits) replaces the FIFO rotation of the standalone workflow.

Commits, in order:

• refactor(ci): Extract a shared fill-release just recipe holding the flags common to all fixture releases (previously inline in build-fixtures); callers append feature params, fork range and output via pass-through args (last flag wins). Fork splitting needs no special handling: --dist loadgroup and testpaths = tests/ are already defaults in pytest-fill.ini.
• feat(ci): The workflow merge described above; deletes nightly-fill.yaml and the fill-nightly recipe.
• chore(ci): Remove a duplicate benchmark entry from evm.yaml (the awk reader matches the first occurrence, yaml.safe_load keeps the last; identical today, a landmine tomorrow).
• refactor(ci): Move the evm override validation from inline bash into generate_build_matrix.py, where the rest of the input validation lives and is unit-tested.
• refactor(ci): Extract the "Check for new commits" inline bash into .github/scripts/check_new_commits.py with unit tests (dispatch short-circuit, baseline bootstrap, new/no commits, gh failure), following the other release helper scripts.

Deliberate trades versus the standalone nightly workflow, flagging for visibility:

• Per-range fixture artifacts are no longer uploaded when a fill leg fails (the composite action only uploads on success); debugging a regressed fill falls back to the job logs.
• A nightly with one failed range publishes nothing (release semantics: combine requires a fully green build); the baseline logic refills everything the next night.
• The cron path cannot run in PR CI, so before merging feat(ci): fill mainnet fixtures nightly, draft cached tests@ releases ethereum/execution-specs#3100 it is worth one workflow_dispatch validation run on a fork, ideally also confirming that the empty-string retention-days on dispatch releases falls back to the repo default as expected.

Net effect on the workflow YAML: the 163-line nightly-fill.yaml is deleted and release_fixtures.yaml grows by ~45 lines of schedule-gated conditionals; the overall diff is dominated by the new script and its tests.

🔗 Related Issues or PRs

Targets the head branch of ethereum#3100; see the review discussion there for context.

✅ Checklist

• All: Ran fast static checks to avoid unnecessary CI fails, see also Code Standards and Verifying Changes:

  just static

• All: PR title have the form <type>(<area>):, where <type> and <area> come from an approrpriate C-<type>, respectively A-<area>, label. The title should match the a target squash commit message.