Implement ConnectionManager — persistent gRPC dial + reconnect loop

[infraax]

Task: ConnectionManager

Parent epic: #4 (M1 — Robot Connection)
Depends on: #14 (SDK), #15 (Registry)

What to do

Create internal/brain/connection.go — wraps one vector-go-sdk connection to one robot. Handles the persistent dial, reconnection on failure, and exposes the SDK client.

type ConnectionManager struct {
    ESN     string
    IP      string
    // ...
}

func NewConnectionManager(esn, ip string, logger *slog.Logger) *ConnectionManager
func (c *ConnectionManager) Start(ctx context.Context) error  // begins connect + reconnect loop
func (c *ConnectionManager) Client() *vectorpb.ExternalInterfaceClient // nil if not connected
func (c *ConnectionManager) IsConnected() bool
func (c *ConnectionManager) State() api.RobotState           // last known state snapshot
func (c *ConnectionManager) Shutdown(ctx context.Context) error

Reconnect logic

• On disconnect: wait 2s, retry. Backoff up to 30s max.
• Log every reconnect attempt at INFO level.
• Never panic on connection failure — always retry.
• One connection per robot. Do not create a second.

Notes

• Use the WirePod source as reference for how it connects to Vector via the SDK.
• The connection uses TLS with certs from WirePod's provisioning — check how WirePod/chipper handles cert loading.
• RESERVE_BEHAVIORS is acquired in a separate task (Acquire and hold RESERVE_BEHAVIORS behavior control #17) after connection is established.

Definition of done

ConnectionManager.Start() dials robot 008093b9, IsConnected() returns true, survives a WirePod restart (reconnects automatically).