Strengthen replan fallback and task tracking

[braedonsaunders]

Summary

• tighten the supervisor replan logic so it never accepts a no-op decision without falling back, records more context (parentTaskId, provider action metadata), and keeps sub-orchestrator child lists in sync
• polish the UI (phase bar, deploy header, orchestrate agent cards) and README to improve usability, visibility, and onboarding guidance
• add a regression test covering resume-from-provider-blocked snapshots to ensure the master orchestrator reopens paused planning tasks

Testing

• Not run (not requested)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 254b05f070

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Run deterministic fallback when replan decision is missing

The new if (!modelDecision) branch exits with action: "accept" before invoking buildDeterministicFallbackDecision(), so any failure path where buildReplanDecision() throws (for example provider/model lookup errors before the structured fallback object is produced) now leaves failed/blocked tasks untouched. That bypasses the retry/skip recovery this commit added for other no-op decisions, and can leave runs stuck with unrecovered failures instead of progressing through deterministic fallback handling.

Useful? React with 👍 / 👎.