fix: add provenance header and speaker IDs to Slack transcript imports

[Kesshite]

Summary

Slack exports are multi-party chats where no speaker is inherently the "user" or "assistant". The parser previously assigned these roles purely by position, allowing a crafted export to place attacker text in the "user" role — making it appear as the memory owner's own words in all future retrieval (data poisoning via stored memory).

This PR adds two mitigations while preserving compatibility with the existing exchange-pair chunking pipeline:

• Provenance header at the top of every Slack transcript: [source: slack-export | multi-party chat — speaker roles are positional, not verified]
• Speaker ID prefix on every message: [U1] Hello instead of just Hello

Role alternation (user/assistant) is preserved so convo_miner.py's > marker-based chunking continues to work — but now every message carries its original speaker identity and the transcript is clearly marked as multi-party with unverified roles.

Refs: #809 (Finding 6)

What changed

mempalace/normalize.py

• _try_slack_json() — provenance header prepended to output; each message prefixed with [speaker_id]

tests/test_normalize.py

• test_slack_json_has_provenance_header — verifies header presence and keywords
• test_slack_json_preserves_speaker_id — verifies [U1], [U2] in output
• test_slack_json_attacker_first_message_attributed — verifies attacker's ID is visible even when placed first

Test plan

• pytest tests/test_normalize.py -v — 91/91 passed
• pytest tests/ -v --ignore=tests/benchmarks — 689 passed, 2 failed (pre-existing version mismatch)
• ruff check — all checks passed
• ruff format --check — all files formatted
• No new dependencies added
• Exchange-pair chunking preserved (user/assistant alternation unchanged)

[bensig]

Code reviewed — no issues found. CLAUDE.md compliance verified.