Pipeline: implement comment-based agent-human communication

Design and implement pipeline communication via task comments.

Current flow: pipeline spawns agent → agent works → commits → done/fail.

Desired flow: 1. Pipeline picks up ready task, reads description + comments 2. Agent reads context. If unclear, posts a comment with questions, sets status to NeedsHelp 3. Human answers via web UI comment 4. Pipeline detects answer, re-spawns agent with full comment history 5. Agent completes work, commits 6. Pipeline verifies, integrates, sets to Done (or Review for human final approval)

Implementation:

• Dev.hs buildDevPrompt already includes task description. Ensure it includes ALL comments (it may already).
• Add a new DevResult variant: DevNeedsInput Text — agent exited cleanly but posted a question comment instead of committing.
• Pipeline sets status to NeedsHelp when agent needs input.
• Pipeline skips NeedsHelp tasks until a new Human comment appears (check comment timestamps vs last agent run).
• The dev workflow prompt (Omni/Ide/Workflows/dev.md) needs instructions: "If the task description is ambiguous or missing critical details, post a comment with your questions using task comment <id> 'your questions' --author agent:engineer --json and exit 0 without committing."

Key files: Omni/Pipeline.hs, Omni/Pipeline/Dev.hs, Omni/Pipeline/Core.hs, Omni/Ide/Workflows/dev.md