feat: support continuous/long-running pipeline execution with automatic issue iteration

[nextlevelshit]

Summary

Enable pipelines to run continuously, automatically picking up and processing new issues one after another without manual re-invocation.

Original Request

i want to be able to run gh-implement for instance for ever and let it one after another open issue implement for instance

Problem Statement

Currently, pipelines execute once and exit. To process multiple issues (e.g., with gh-implement), the user must manually re-run the pipeline for each issue. This is tedious for batch processing workflows.

Proposed Behavior

• A pipeline can be invoked in continuous mode (e.g., wave run --continuous gh-implement)
• In continuous mode, after completing one iteration the pipeline:
  1. Polls for the next available issue (using configured filters/labels)
  2. Creates a fresh workspace and executes the pipeline for that issue
  3. Repeats until no more matching issues exist or the user interrupts (Ctrl+C)
• Each iteration is fully isolated (fresh memory, fresh workspace) per Wave's existing contract
• Graceful shutdown on interrupt — completes current step, then exits

Acceptance Criteria

• Pipeline manifest supports a continuous or loop execution mode
• wave run --continuous <pipeline> iterates over matching issues automatically
• Each iteration gets a clean workspace and fresh agent memory
• Graceful shutdown on SIGINT/SIGTERM (finish current step, then stop)
• Progress events emitted for each iteration (issue number, status)
• Pipeline exits cleanly when no more matching issues are found
• Documentation updated with continuous mode usage

Technical Considerations

• How does the pipeline determine which issue to process next? (label filter, oldest first, priority?)
• Should there be a configurable delay between iterations?
• How to handle failures — skip and continue, or halt?
• Rate limiting considerations for GitHub API calls
• State tracking to avoid re-processing already-completed issues