Expose enhancement availability through the live session

## Problem

Frontends cannot determine whether the active set profile contains confirmed, compatible AI-assisted semantic enhancement. Inferring this from profile maturity or contextual-evidence status would conflate empirical, deterministic, and AI-assisted evidence.

## Proposed change

Add immutable enhancement availability state to the shared live-session snapshot.

Represent at least enhanced and available; not enhanced; incompatible or invalid enhancement; unavailable profile; and user-disabled AI-enhanced suggestions.

Expose an explicit command for enabling or disabling AI-enhanced suggestions. The session enforces availability: an unenhanced profile cannot activate the mode.

Keep all profile interpretation and gating in Python. Frontends only render state and send commands.

## Acceptance Criteria

- [ ] AC1: `tests/test_session.py` verifies every enhancement availability state from representative profile load outcomes.
- [ ] AC2: Enhanced availability requires a compatible confirmed profile; profile maturity alone is insufficient.
- [ ] AC3: Enabling the mode for an unenhanced or incompatible profile leaves it disabled and publishes an explanatory state.
- [ ] AC4: Profile changes atomically update availability and cannot leave an enabled mode attached to an unenhanced set.
- [ ] AC5: Recommendation snapshots distinguish existing contextual evidence from AI-enhanced relationship availability.
- [ ] AC6: `tests/test_mock_session.py` exposes deterministic representative states through the same immutable contract.

## Dependencies

- Apply enhanced relationships to draft scoring.

## Implementation classification

- **Estimated size:** S
- **Orchestration risk:** Normal
- **Reason:** This is a focused shared-state and command-boundary addition.

Start by reading and running tests/test_session.py and tests/test_mock_session.py to understand the immutable live-session contract and representative profile outcomes. Trace the existing session commands and recommendation snapshots before implementing the availability states and enable/disable behavior. Done means all acceptance criteria pass, including atomic profile updates and distinct contextual versus AI-enhanced availability.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.