Bug: taskctl pipeline - no PM escalation after 3 adversarial rejections causes infinite loop

[randomm]

What happened

When the adversarial-pipeline rejects a developer implementation 3+ times, the pipeline does not escalate to the PM. Instead, it keeps respawning the developer indefinitely until the thread dies (context limit or token budget exhausted).

Expected behaviour

After 3 adversarial rejections (ISSUES_FOUND or CRITICAL_ISSUES_FOUND), the pipeline should:

1. Stop respawning the developer
2. Escalate to PM with verdict summary
3. Allow PM to decide: override, re-scope, or abort

Current behaviour

• No maximum retry limit configured
• No escalation logic in pulse.ts
• Pipeline loops forever: developer → adversarial → respawn → repeat
• Thread dies when LLM context limit reached

Steps to reproduce

1. Run taskctl start on an issue with complex requirements
2. Wait for adversarial to find issues in round 1
3. Developer fixes, adversarial finds NEW issues (or same issues reworded)
4. Repeat 3+ times
5. Observe: No PM escalation, pipeline continues indefinitely

Environment

• Branch: dev
• Files involved: src/tasks/pulse.ts, src/tasks/pulse-scheduler.ts, src/tasks/pulse-verdicts.ts

Acceptance Criteria

• Pipeline tracks pipeline.attempt counter for each task
• After 3 adversarial rejections, pipeline escalates to PM
• PM receives notification with verdict history and issue summary
• Pipeline pauses, waiting for PM decision (override/re-scope/abort)
• Test added verifying escalation after 3 retries

Definition of Done

• Escalation logic added to pulse.ts or pulse-scheduler.ts
• PM notification mechanism implemented (comment on issue, or session message)
• Test written verifying escalation after 3 rejections
• Documentation updated in AGENTS.md describing escalation behavior

[randomm]

Fixed in #389