Add bounded synchronous runner for source and flow exploration

[chubes4]

Problem

Local exploration still falls back to direct handler calls or wp eval because normal flow execution is asynchronous and Action Scheduler-oriented. That is correct for production, but awkward when debugging or exploring a source/brain pipeline interactively.

We need a bounded synchronous runner that executes a flow/pipeline/source step and prints packets/diagnostics without requiring manual Action Scheduler drains.

Desired CLI

wp datamachine flow run-sync <flow-id> --max-steps=20 --show-packets --format=json
wp datamachine pipeline run-sync <pipeline-id> --input-file=... --show-packets --format=json

Desired behavior

• Execute bounded steps inline for local exploration.
• Stop cleanly at --max-steps, --max-items, or timeout.
• Include packet counts, handler configs, tool params, errors, and per-step output summaries.
• Never replace production Action Scheduler execution; this is an explicit debug/exploration mode.

Why

The Trac Report 43 session needed repeated source calls and packet inspection. The lower-level tools worked, but the production flow runner was too indirect for tight iteration.

Out of scope

• Background scheduling changes.
• Changing normal flow run behavior.

AI assistance

• AI assistance: Yes
• Tool(s): OpenCode (GPT-5.5)
• Used for: Drafted from live Data Machine/Intelligence source exploration friction.