[pipelines] Pipeline definition format + durable runner (trigger → declarative pipeline)

[serge-ivo]

Child of #94. Builds on #92 (trigger action), #52 (durable execution substrate).

Define the pipeline as data and a runner that executes it durably.

Do

• A JSON pipeline schema: ordered steps[] (each {tool, inputs, bind}), typed inputs from settingsSchema + trigger payload, a declared sink collection.
• Store pipeline definitions per agent/instance (data, not the closed workflow union — see Triggers: let a cron/webhook trigger run a durable pipeline, not just the 4 canned actions #92).
• A durable runner (escapes the 30s DO limit; resumable — same machinery as JOB_APPLY/CODING_SESSION) that walks the steps, passing outputs between them, honoring pagination/fan-out + concurrency caps.
• Invocable by: a cron/webhook trigger (Triggers: let a cron/webhook trigger run a durable pipeline, not just the 4 canned actions #92) and an LLM-callable run_pipeline tool ([connectors] Generic tool-call API + MCP auto-generation #87) so an agent can start it on request (e.g. chat: "sweep Sydney").

Acceptance

• A cron trigger runs a declared pipeline that reads params from settings and writes to a collection.
• "Run pipeline with param city=Sydney" works from chat and from MCP, resumably.