Bulk-backfill _summary on pre-existing components (app-context)

[taylortom]

Follow-up to #138 / PR #139.

_summary is computed at write time, so components get a summary the next time they are edited. Components that predate the feature stay empty until then.

Add an app-context operation that walks every existing component, recomputes _summary via ContentModule.computeSummary, and writes it back in bulk — to light up the back-catalogue without requiring an edit per component. This can't be a migration: migrations run during boot with only a raw db handle, and summary computation needs the resolved per-component schema (module must be ready).

Likely shape: a ContentModule.reindexSummaries() method triggered by a bin/ script or an authenticated admin endpoint (write:content/super). Not blocking the feature.

[taylortom]

Decision: run this on-demand, not at boot.

Rationale:

• It is a one-off catch-up for pre-existing content — write-time compute keeps everything current from here on, so the bulk pass only needs to run once per deployment of the feature. Running it every boot is wasted work.
• Avoiding re-runs would require a "have I done this?" guard flag — i.e. re-implementing the version bookkeeping a migration gives for free, which is the very thing we cannot use here (migrations run pre-module-ready and deadlock; see New: compute and expose per-component _summary #139).
• It would delay onReady/serving on the boot it runs (bulk sweep + writes over potentially thousands of components).
• In a multi-instance/HA deploy every instance would fire it at boot simultaneously — redundant writes and races unless we add leader-election/locking.

Preferred shape: an explicit operator-triggered operation — a bin/ script or an authenticated admin action (write:content/super) calling ContentModule.reindexSummaries() — giving observable output, control over timing (run off-peak), no startup penalty, and no locking. A guarded run-once-at-boot is possible but only worth it if invoking a post-deploy command is a real problem in the ops flow.