feat: add explicit decision relations to FLG

[dlxeva]

Summary

This PR adds a small, file-backed decision relation layer to FLG.

• adds four explicit relation types: supersedes, supports, conflicts_with, and depends_on
• adds relation flags to flg decision add
• rejects malformed CLI IDs and references to decisions missing from the formal ledger
• preserves valid IDs while reporting malformed values in human-edited relation declarations
• shows outgoing, incoming, and malformed relation data in flg trace
• reports malformed values, unknown targets, and self-references in flg doctor
• adds bilingual Markdown parsing, focused tests, and a v0 design note

Why

FLG already has durable decision IDs, provenance episodes, and a human-readable ledger. This adds explicit links between judgments without introducing a graph database or a second source of truth.

D-006 --supersedes----> D-004
D-006 --supports------> D-003
D-006 --conflicts_with> D-005
D-006 --depends_on----> D-001

Design boundaries

• DECISIONS.md remains the source of truth.
• Relations are parsed deterministically from Markdown.
• Superseded decisions remain in history.
• A relation does not automatically mutate decision status.
• Decision Relations v0 targets existing D-* formal decisions only.
• Broader dependencies and derived_from remain conceptual or evidence-provenance relationships in v0.
• No LLM inference, embeddings, graph database, RDF, or new runtime dependency is introduced.

Example

flg decision add \
  --decision "Ship the manifest-first context mode" \
  --rationale "It preserves continuity under bounded context budgets" \
  --supersedes D-004 \
  --supports D-003 \
  --conflicts-with D-005 \
  --depends-on D-001,D-002

flg trace D-006
flg doctor --strict

Validation

• 198 tests passed locally
• repository smoke test passed
• git diff --check passed
• CI passed on Python 3.10, 3.11, and 3.12

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 109e39e45d

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]

Report malformed relation values instead of dropping them

When the human-editable ledger contains a mistyped relation such as - **Depends On:** D-99O, _extract_ids returns an empty list, so both flg trace and flg doctor --strict treat the declaration as if no relation existed. Because DECISIONS.md is authoritative and may be edited outside decision add, parsing should preserve or report non-empty malformed relation values rather than silently discarding them.

Useful? React with 👍 / 👎.