Communication protocol should track active requests

The communication protocol ought to have an ID associated with it so we can cancel DAGs mid-execution.

Workflow might look something like this:
1. Creates ID for new execution
2. Adds ID to active requests
3. Runs DAG - each node type should make some check to ensure the request ID is still in the active requests
4. When DAG finishes, remove ID from active requests

A user could then submit a cancel request that removes the ID from the active requests pool. The next node that checks the active pool will see the ID is no longer around and stop execution. This is especially useful for long-running executions, such as querying an LLM.

No files, tests, or concrete entry points are named. Start by locating the communication protocol and DAG execution paths, then trace how executions and long-running LLM work are represented. Done should cover request IDs, active-request tracking, cancellation requests, and checks across node types, but the issue does not define the required interfaces or tests.

Written by the indexing model from the issue text.