[Server] Implement Server Lite Mode (gRPC Ingest + WS Broadcast)

[varev-dev]

Goal

Implement a Server Lite Mode that accepts agent metric batches over existing gRPC ingest and immediately broadcasts normalized live updates to connected dashboard clients over WebSocket.

Scope of work

• Keep gRPC ingest endpoint compatible with current protocol.
• Validate incoming batches using existing core invariants:
  • len(timestamps) == len(values)
  • strictly increasing timestamps per TimeSeries
  • metric name format and basic batch limits.
• Track latest in-memory state per agent/metric needed for live dashboard.
• Expose WebSocket endpoint for live stream broadcast to UI clients.
• Broadcast events to all subscribed clients with minimal transformation (agent id, metric name, labels, timestamp, value).
• Handle disconnect/reconnect gracefully for both agents and WS clients.
• Add observability hooks/logging for ingest throughput, connected WS clients, and dropped/invalid messages.

Acceptance criteria

1. Agent can complete handshake and stream metric batches to server.
2. Server responds with ack and correct sequence numbers.
3. Invalid batches are rejected/logged and do not crash the process.
4. At least one WS client can connect and receive live metric updates within acceptable real-time delay.
5. Multiple concurrent WS clients receive the same stream reliably.
6. When no WS clients are connected, ingest remains stable (no blocking/backpressure failure).
7. Basic tests cover:
   • gRPC ingest,
   • WS broadcast,
   • validation failures.

Deliverables

• Ingest-to-broadcast pipeline implementation.
• WebSocket live endpoint contract and message schema documentation.
• Unit/integration tests.
• Update to server README/docs with:
  • how to run the server,
  • endpoint list,
  • known limitations (no persistence/history).