docs: clarify HubTargets.mid is informational-only, not benchmarked (#2039)

mid is computed by selectHubTargets() and covered by tests, but neither
benchmark.ts nor query-benchmark.ts passes it into any benchDepths/benchQuery
call — it only ever reaches a console.error log line and the output JSON's
targets block. This predates PR #2039 (the original pre-extraction
selectTargets() had the same shape), so wiring it into an actual measurement
is a distinct feature change with its own design questions (what should it
measure, does the benchmark-gate baseline config need a new entry), not a
determinism/kind-filter bug fix.

Document the current informational-only status directly on the interface
field so it isn't mistaken for an active measurement. Tracked in #2089 for
whoever wants to either wire it in or drop it.

Author: carlos-alm

scripts/lib/hub-selection.ts

@@ -43,6 +43,13 @@ export const PINNED_HUB_CANDIDATES: readonly string[] = ['buildGraph', 'openDb',
 export interface HubTargets {
 	hub: string;
 	hubFile: string;
+	/**
+	 * Node at the median edge-count rank among kind-filtered candidates.
+	 * Informational only: it is recorded in benchmark output for visibility
+	 * but neither `query-benchmark.ts` nor `benchmark.ts` currently passes it
+	 * into a `benchDepths`/`benchQuery` call, so it drives no timed
+	 * measurement. See #2089 to wire it into an actual measurement or drop it.
+	 */
 	mid: string;
 	leaf: string;
 }