refactor(global-db): run configuration and rollup writes as jobs

[ScriptedAlchemy]

Step 5i-3 of the storage plan: global-db gets RegisteredGlobalDb::transact and its configuration and observability rollup writes become writer jobs.

API. RegisteredGlobalDb::transact(body) runs body as one job over Database::transact and returns Result<Result<R, E>>. The body gets a RegisteredGlobalDbWriteTx, which implements QueryExecutor and Executor, so the existing &impl Executor helpers run in it unchanged. The write scope is checked before queueing. The runtime verifies the same authority at COMMIT, so a scope that ends mid-body rolls the job back. RegisteredGlobalDbWriteTx is exported (approved).

Sites converted. Configuration store (initialize, daemon source binding, registry-shape convergence, component activation, profile commit_direct), control (save_plan, commit_direct, commit_protected, dry_run_rollback, apply_rollback), revision (save_change_plan, commit), and the rollup rebuild, dirty-day claim and release, empty-day frontier initialize, claim and release, and compaction. The configuration mutation-scope probe is now a direct require_active_write_scope check instead of begin-and-rollback.

Behavior kept. A rollup rebuild that finds its stored receipt still ends by rolling back (RebuildAbort::Replayed), so replay does not advance graph_transaction_generation. Bodies are Fn, so each site clones its inputs into the closure. Job failures report one failed to <operation> error where the interactive path reported begin, commit and close separately.

Proof. transact_rolls_back_when_scope_ends_before_commit and copied_transact_handle_cannot_write_after_scope_drop_inside_the_body cover the primitive. A mutation that removed the wrapper's own commit-time scope check left both green, because the runtime already verifies authority at commit. The check is deleted and the decision recorded in decisions.tsv.

Verification (Bazel). global-db unit_test and its five integration targets, rustfmt_check_2024 and --config=clippy on global-db, tracedecay-application unit_test and application_suite, and tracedecay daemon_suite, mcp_suite and session_suite pass. session_suite failed once with 39 schema-init lease timeouts while the host load average was near 500 and passed on a rerun.

Not in this PR: the other global-db sites (stack delivery, observation projection, analytics, LCM and the rest), the old RegisteredGlobalDbWriteTransaction, and the interactive API deletion.

---

[changeset-bot]

⚠️ No Changeset found

Latest commit: 52b02a9

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.