Investigate + close the config-as-code gap for impactMap

[JSONbored]

Part of #4092.

Confirmed finding (was originally filed as an open investigation — now resolved by direct code-check)

impact-map-wire.ts:30: return isImpactMapEnabled(env) && manifestImpactMapEnabled; — review.impactMap IS already consulted for the real on/off decision, but only as a REQUIRED additional AND-gate alongside GITTENSORY_REVIEW_IMPACT_MAP, same pattern as inlineComments/fixHandoff (#4099) and reviewMemory (#4101) — not the fully-dead features.* pattern this epic also found elsewhere. A repo can restrict via config but can't fully replace the env var.

Fix

Same shape as #4099/#4101: apply src/review/feature-activation.ts's existing precedence so review.impactMap: true can turn the feature on for a repo even without the env var/allowlist, and false forces it off. Follow whatever namespace decision #4099 makes for consistency.

Deliverables

• isImpactMapEnabled's real gate migrated to the resolver's full precedence
• Regression test: unset config preserves exact current behavior
• Tests mirroring the feature-activation pattern

Expected outcome

review.impactMap fully controls the feature per-repo, consistent with #4099/#4101.

Effort

S — confirmed, well-defined fix; no further investigation needed.

[linear-code]

JSO-3152

[JSONbored]

Closing as investigated, no code change warranted — same finding as #4101 (see its closing comment for the full evidence), which applies identically here.

impact-map-wire.ts's isImpactMapEnabled(env) && manifestImpactMapEnabled was never a "required-AND-because-allowlist" shape like inlineComments/fixHandoff's pre-#4116 code — there is no isConvergenceRepoAllowed term anywhere in impactMap's history (confirmed: impactMap is absent from feature-activation.ts's FEATURE_GLOBAL_FLAG map). resolveReviewPromptOverrides (src/signals/focus-manifest.ts) already collapses impactMap to a definite boolean — manifest?.review.impactMap === true — before it ever reaches the AND-gate, so envFlag && definiteBoolean is algebraically identical to any "kill-switch, then let the manifest decide" rewrite for every input. review.impactMap already fully controls this feature today (subject to the operator kill-switch, same as every sibling toggle) — there was nothing to bypass.

PR #4125 (opened against this issue) is a byte-identical code-shape change with no behavior delta. Closing unmerged rather than shipping a "fix" for a gap that doesn't exist, consistent with #4101's disposition.