feat: org connection limits and FIFO queue in cluster mode

[EDsCODE]

Overview

This PR adds per-org connection limits for Duckgres cluster mode with FIFO queueing. When an org is at its configured max_connections, new session requests wait in that org's in-memory FIFO queue for up to worker_queue_timeout, which now defaults to 60s. If a request is still queued when the timeout expires, startup fails with SQLSTATE 53300 and message too many connections.

Proposed Changes

• Config Store Schema: Adds max_connections to Org and carries it into OrgConfig snapshots.
• Per-Org FIFO Admission Queue:
  • SessionManager tracks active slots and queued waiters per org.
  • Over-limit requests are admitted FIFO when sessions end, worker crash cleanup releases slots, failed session creation releases a reserved slot, or max_connections increases.
  • Queued client cancellations return cancellation, while queue deadline expiry returns ErrTooManyConnections.
• Timeout Behavior:
  • Reuses existing worker_queue_timeout for worker acquisition and per-org connection-limit queue admission.
  • Changes the default from 5m to 60s and documents it in CLI help, README, and example config.
• Database-Native Error Pattern:
  • Over-limit timeout maps to SQLSTATE 53300 with message too many connections.
• Admin API:
  • Exposes max_connections in UpdateOrg.
  • Preserves existing org fields when updating only max_connections.
• Worker Context Fix:
  • Ensures child worker connections have a connection context before sending initial parameters.

Verification

• go test -count=1 ./controlplane -run 'TestSessionManager_ConnectionLimits'
• go test -count=1 . -run TestResolveEffectiveConfigWorkerQueueTimeoutDefault
• go test -count=1 .
• just test-controlplane
• just test-controlplane-k8s
• git diff --check
• just lint attempted; blocked locally because golangci-lint is not installed on PATH.