feat(pipeline): add multi-step agent pipeline execution

[garethpaul]

Chains agent spawns where output flows between steps automatically. You write a YAML file, each step spawns an agent, and later steps can reference earlier step output via
{{steps..output}}.

Example yaml

name: research-fix-review
 description: Research codebase, fix issues, then review changes

 defaults:
   agent: claude-code
   mode: read

 steps:
   # Step 1: Research — runs first, uses defaults (claude-code, read mode)
   - name: research
     prompt: |
       Analyze this codebase and identify:
       1. Security vulnerabilities
       2. Performance bottlenecks

   # Step 2: Two fixes run concurrently after research completes
   - parallel:
       - name: fix-security
         agent: codex
         mode: edit
         model: o3-pro
         prompt: |
           Fix the security issues found below. Only fix security, nothing else.

           {{steps.research.output}}

       - name: fix-performance
         mode: edit
         prompt: |
           Fix the performance issues found below. Only fix performance, nothing else.

           {{steps.research.output}}

   # Step 3: Final review — runs after both parallel steps complete
   - name: final-review
     prompt: |
       Review the changes that were just made:

       Security fixes:
       {{steps.fix-security.output}}

       Performance fixes:
       {{steps.fix-performance.output}}

       Did the fixes introduce any new issues?

• Add pipeline feature for chaining agent calls where output flows between steps automatically via `{{steps..output}}` interpolation
• YAML schema supports sequential steps, parallel groups, defaults inheritance, and per-step overrides (agent, mode, model, cwd, args)
• CLI: `poe-code pipeline run `, `poe-code pipeline validate `, `--dry-run` support
• SDK: `runPipeline()` and `parsePipeline()` exported for programmatic use
• unit tests across parse, validate, interpolate, run, and CLI command modules

Test plan

• 15 parse tests (valid YAML, defaults, parallel groups, all error cases)
• 13 validate tests (agent resolution, forward refs, parallel sibling refs, unknown refs)
• 9 interpolate tests (step output/exitCode, pipeline vars, multiline, unresolved refs)
• 10 run tests (sequential, parallel, interpolation, abort on failure, defaults, args, cwd)
• 5 CLI command tests (validate valid/invalid, run spawns, dry-run, failure exit code)
• All 2502 repo tests passing
• Lint clean"