fix(dashboard): accept plain dict in agent_run for V2 executions

[petarkrvavac]

Problem

Starting an autonomous agent run from the V2 dashboard failed immediately with:
AttributeError: 'dict' object has no attribute 'model_dump'
The direct POST /api/agent/run route worked, but V2 executions failed before the first agent round.

Root cause

agent_run() assumed that body was always a Pydantic AgentRunRequest and unconditionally called:
body.model_dump(exclude_defaults=True)
That is true for the direct FastAPI route, where the request body is validated into AgentRunRequest.
The V2 execution engine uses a different path: _agent_execution() calls agent_run(args) with a plain dict, which does not provide .model_dump().

Fix

Normalize the body at the start of agent_run():

• use model_dump(exclude_defaults=True) when available
• otherwise copy the plain dict
  This keeps the direct route behavior unchanged while allowing V2 executions to use the same handler safely.

Verification

• Direct POST /api/agent/run behavior remains unchanged
• V2 autonomous runs no longer fail on model_dump()
• python -m py_compile wallbreaker/dashboard/server.py succeeds