fix(server): NET-6719 mark plan-approval retry runs as safe bootstrap

The reaper's plan-approval continuation retry path runs after the same
CAS write that terminalized the run as process_lost. The shared
scheduleBoundedRetryForRun helper still applies the legacy
legacyExecutionNeedsReconciliation gate, which refuses to schedule a
retry when the failed run lacks explicit bootstrap evidence.

The c8d2b2fd0 fix-forward added that evidence to every reaped run, but
the follow-up 035f4dd79 commit removed it again so monitor-dispatch
losses would still escalate to the board instead of generating spurious
process_lost_retry rows.

Keep both behaviours: only the interaction_continuation_infra_retry
path (which has its own eligibility gates via
isResolvedInteractionContinuationWakeContext and
isRetryableInteractionContinuationInfrastructureFailure) needs the
bootstrap evidence. Add it on the in-memory copy passed to
scheduleBoundedRetryForRun inside
scheduleInteractionContinuationInfrastructureRetryIfEligible so the
shared gate sees safe bootstrap while the persisted row stays
unchanged for the monitor-dispatch / process_lost_retry paths.

NET-6719 / NET-6853

server/src/services/heartbeat.ts

@@ -15830,7 +15830,21 @@ export function heartbeatService(
       return null;
     }
 
-    return scheduleBoundedRetryForRun(run, agent, {
+    // The reaper promoted this run to a process_lost CAS failure before any
+    // provider work produced output; the retry is an explicit
+    // infrastructure-loss replay, not an ambiguous bootstrap that the legacy
+    // reconciliation gate is meant to block. Mark the failed run as safe
+    // bootstrap evidence on the in-memory copy we hand to
+    // scheduleBoundedRetryForRun so the shared gate does not refuse the retry.
+    const runForRetry: typeof run = {
+      ...run,
+      resultJson: {
+        ...(parseObject(run.resultJson) ?? {}),
+        executionRecovery: { kind: "bootstrap", providerWorkStarted: false },
+      },
+    };
+
+    return scheduleBoundedRetryForRun(runForRetry, agent, {
       retryReason: INTERACTION_CONTINUATION_INFRA_RETRY_REASON,
       wakeReason: INTERACTION_CONTINUATION_INFRA_WAKE_REASON,
       maxAttempts: INTERACTION_CONTINUATION_INFRA_MAX_ATTEMPTS,