Implement Matrix adapter with E2EE posture and bridge metadata preservation

[mdheller]

Agent work order

@copilot @codex please implement this as a focused adapter lane after PR #1 lands or by branching from seed/agentterm-chatops-core if instructed.

Goal

Implement the first live Matrix adapter for AgentTerm.

Context

AgentTerm is Matrix-first. Slack/Discord are bridge targets, not the canonical substrate. The adapter must preserve Matrix event metadata rather than flattening room messages into plain chat text.

Required scope

• Read/write Matrix room events behind the AgentTermAdapter boundary.
• Preserve room ID, alias, event ID, sender MXID, thread/root event where available, membership changes, redactions, bridge metadata, and encryption state.
• Surface encrypted-room posture before any sensitive context is released to agents.
• Store normalized events through EventStore.
• Add config loading from configs/agent-term.example.json shape or a minimal compatible subset.
• Add tests using fake Matrix payloads; do not require a live homeserver in CI.

Non-goals

• Do not implement Slack/Discord first.
• Do not hardwire Matrix SDK objects into cli.py.
• Do not leak encrypted-room content when posture is unknown.

Validation

ruff check .
pytest