propose: FEIGN_CLIENT role → CLIENT + HTTP_CLIENT capability (#28)

* propose: FEIGN_CLIENT role → CLIENT + HTTP_CLIENT capability

Add CLIENT-ROLE-RENAME-PROPOSE.md:
- Rename role enum FEIGN_CLIENT → CLIENT (Spring-native vocabulary)
- Add HTTP_CLIENT capability (peer to MESSAGE_PRODUCER which already exists)
- Keep auto-promotion scoped exactly as today (only @FeignClient interfaces)
- RestTemplate/WebClient users opt in via brownfield @CodebaseRole annotations
- Two-phase deprecation alias: FEIGN_CLIENT accepted as alias for one release
- Ships AFTER PR-F1 and PR-G1; ontology bump 8→9 (or 7→8 if order changes)

Delete DEFERRED-REST-CLIENT-MIGRATION-PROPOSE.md:
- Superseded by the new propose
- Its 2026-04-26 conclusion (REST_CLIENT as capability, not role rename)
  is reversed by this proposal — the architecture has moved on:
  capabilities already exist, brownfield overrides already exist, and
  @CodebaseRole already lets users assign primary roles honestly.

* propose: drop two-phase deprecation alias (hard rename in v1)

MCP bundle has no users yet — breaking changes are explicitly allowed.
Simpler, smaller, cleaner:

- Remove FEIGN_CLIENT from VALID_ROLES in the same PR (no alias)
- Brownfield YAML/annotation inputs that still say FEIGN_CLIENT fail
  validation via the existing VALID_ROLES check; error message lists
  allowed roles, redirecting users to CLIENT + HTTP_CLIENT
- No new helper code in graph_enrich.py (was ~15 LOC alias machinery)
- LOC estimate ~120 → ~105
- Test #4 flipped from 'alias translated' to 'role rejected'
- §9 [TBD] reduced from 7 → 5 (alias-specific items removed)
- Risks table updated (row 1 likelihood N/A; old row 6 'alias complexity'
  removed)

Author: HumanBean17