Support concurrent heartbeat runs with maxConcurrentRuns policy

Add per-agent maxConcurrentRuns (1-10) controlling how many runs
execute simultaneously. Implements agent-level start lock, optimistic
claim-then-execute flow, atomic token accounting via SQL expressions,
and proper status resolution when parallel runs finish. Updates UI
config form, live run count display, and SSE invalidation to avoid
unnecessary refetches on run event streams.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ui/src/components/ActiveAgentsPanel.tsx

@@ -155,7 +155,6 @@ export function ActiveAgentsPanel({ companyId }: ActiveAgentsPanelProps) {
   const { data: liveRuns } = useQuery({
     queryKey: queryKeys.liveRuns(companyId),
     queryFn: () => heartbeatsApi.liveRunsForCompany(companyId),
-    refetchInterval: 5000,
   });
 
   const runs = liveRuns ?? [];

ui/src/components/NewAgentDialog.tsx

@@ -112,6 +112,7 @@ export function NewAgentDialog() {
           intervalSec: configValues.intervalSec,
           wakeOnDemand: true,
           cooldownSec: 10,
+          maxConcurrentRuns: 1,
         },
       },
       budgetMonthlyCents: 0,

ui/src/components/OnboardingWizard.tsx

@@ -154,6 +154,7 @@ export function OnboardingWizard() {
             intervalSec: 300,
             wakeOnDemand: true,
             cooldownSec: 10,
+            maxConcurrentRuns: 1,
           },
         },
       });

ui/src/components/agent-config-primitives.tsx

@@ -36,6 +36,7 @@ export const help: Record<string, string> = {
   graceSec: "Seconds to wait after sending interrupt before force-killing the process.",
   wakeOnDemand: "Allow this agent to be woken by assignments, API calls, UI actions, or automated systems.",
   cooldownSec: "Minimum seconds between consecutive heartbeat runs.",
+  maxConcurrentRuns: "Maximum number of heartbeat runs that can execute simultaneously for this agent.",
   budgetMonthlyCents: "Monthly spending limit in cents. 0 means no limit.",
 };