fix: isolate WordPress runtime state per execution test

[JasonTheAdams]

Why this matters

WP-Bench scores are only meaningful if they're reproducible. Today, execution tests run concurrently against one shared WordPress instance with no reset between tests. WordPress is deeply stateful — options, posts, roles, hooks, transients, cron events all persist in the database — so code generated for one test can silently change the outcome of the next. In multi-model mode the problem compounds: one model's leftover state can help or hurt the next model's scores. That means rankings can shift based on test ordering and scheduling luck, and a provider who re-runs the suite may get different numbers. No leaderboard can be trusted on that foundation. This change makes every execution test start from a known-clean WordPress baseline, which is the prerequisite for every scoring and auditability improvement that follows.

Changes

• RunConfig: new execution_isolation (reset_per_test default | none) and execution_concurrency (default 1). A model validator rejects execution_concurrency > 1 under reset_per_test with a clear error instead of silently sharing state.
• core.py: single-model, multi-model, and reference-solution execution paths now route through one shared _run_isolated_execution_loop() that calls environment.reset() before every test. Knowledge tests keep their existing concurrency (they never touch the WordPress runtime).
• WordPressEnvironment.reset(): wp db reset drops all tables and leaves WordPress uninstalled, so reset now follows with wp core install to return to a deterministic just-installed baseline (grader.base_url added for the install URL).
• Result metadata: metadata.runtime_isolation records the mode used, in both single- and multi-model payloads, so any results file shows how it was produced.
• Docs: README + wp-bench.example.yaml document the isolation semantics and the official-run requirement.

Verification

• pytest python: 41 passed (8 new isolation tests: reset-before-every-test in all three modes, reset between models, concurrency rejection, metadata recording, legacy opt-out)
• ruff check python: clean
• mypy python: 3 pre-existing trunk errors only (none introduced)
• Runtime smoke test vs Docker/wp-env: not run in this PR (no local runtime up); the reset→reinstall sequence follows the wp-env docs and a state-leak interleaving assertion is covered in the mocked tests.

Notes for reviewers

• First PR in a series moving the harness toward provider-grade reproducibility; subsequent PRs stack on this branch.
• A per-worker isolated-runtime pool (parallel execution with full isolation) is intentionally deferred to keep this diff reviewable — the execution_isolation config surface leaves room for a future container_per_worker mode.

[github-actions]

The following accounts have interacted with this PR and/or linked issues. I will continue to update these lists as activity occurs. You can also manually ask me to refresh this list by adding the props-bot label.

If you're merging code through a pull request on GitHub, copy and paste the following into the bottom of the merge commit message.

Co-authored-by: JasonTheAdams <jason_the_adams@git.wordpress.org>
Co-authored-by: lezama <migueluy@git.wordpress.org>

To understand the WordPress project's expectations around crediting contributors, please review the Contributor Attribution page in the Core Handbook.

[lezama]

rebased on trunk + green, one fix for #24's required test_function field 👍