refactor(global-db): run more registered writes as job transactions

[ScriptedAlchemy]

Part of #4099 (storage redesign, cutover step 5, slice 5i-4). Does not close it.

What changes

About twenty more global-db interactive write transactions now run as RegisteredGlobalDb::transact jobs, so each is one bounded writer job instead of a held transaction.

• transact_or(wrap, body) maps a refused or failed job to the site's own error type. transact_or_rollback(wrap, body) with JobExit::{Rollback, Fail} lets a body end early with a value the caller wants as success (a replayed append). The job rolls back, so a replay leaves no commit. transact_config and transact_rollup now use transact_or.
• Converted: accounting (2), workflow index (2), git topology anchor (1), GitHub stack delivery (4), delivery settlement (2), source rewrite presence, begin and complete (3), registry forget and retire (2) plus the shared delete helper, remote deletion (3), discovery queue (1), analytics append, outbox claim, delay and settle, retention prune, and append with cursor.
• Helpers under these sites take &impl Executor or &impl QueryExecutor, so the same body serves the job and the sites still on interactive transactions.
• WorkflowIngestWriteTxn (sessions trait) and its global-db impl are deleted; nothing used them after the workflow adapter moved.

Bodies are Fn; each site clones what it owns into the closure and none has an effect outside storage.

Left for a later slice

• apply_registry_gc reads outside the transaction while holding it.
• relink_orphan_store_instance has a manifest file effect between the SQL and the commit.
• claim_and_settle_observability_emissions takes an FnMut that allocates producer sequences; a re-run would allocate twice.
• append_analytics_event uses the writer connection, not an interactive transaction.
• Unread groups: project_registry.rs, observation_projection, registered_lcm*, registered_provider_usage, registered_dashboard, registered_sessions, git_correlation_adapter, schema_stages and schema_contract invariants, git_index_transactions/database.rs, observation/retention.rs, observation/canonical_bodies.rs, observation/schema.rs.

Evidence

• New tests in tests.rs next to the 5i-3 job tests: transact_or_rollback_keeps_an_early_value_but_never_a_commit, transact_or_wraps_a_scope_that_ended_mid_body_and_rolls_back, converted_write_is_refused_without_a_write_scope.
• Existing tests that now exercise converted multi-write sites, including analytics_import_cursor_failure_rolls_back_events and analytics_import_cursor_conflict_rolls_back_events, plus the stack delivery, settlement, source rewrite and outbox suites.
• Bazel: global-db unit_test and registered_schema_fail_closed, sessions unit_test, //crates/tracedecay:unit_test, :daemon_suite, :mcp_suite, :session_suite all pass; --config=clippy on global-db and sessions clean; rustfmt_check_2024 passes for both.
• No mutation run on the new helpers: the rollback is the runtime's, and the tests assert its effect through the new entry points.

---

[changeset-bot]

⚠️ No Changeset found

Latest commit: 19dd56d

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.