fix(rust_brain): preserve HLC on snapshot restore and gossip receive

[cursor]

Bug and impact

After v0.6.0 added HLC monotonic enforcement, restore_from_file() and gossip.receive() still assigned fresh HLC timestamps instead of preserving stored values.

Concrete scenario: Node A writes k=v1 with HLC (5000, 10, nodeA), snapshots, and restores after crash. A causal successor write k=v2 with HLC (5000, 11, nodeA) raises TimestampRegression and is silently dropped — post-disaster recovery loses writes. Gossip without HLC could also overwrite fresher data on existing keys.

Root cause

• restore_from_file() rebuilt MemoryNode without the hlc field and never advanced the global HLC clock.
• bulk_write() and gossip.receive() did not pass through HLC from wire payloads.

Fix

• Add _parse_hlc() for snapshot/wire normalization (with legacy ts_ns fallback for pre-v0.6.0 snapshots).
• Restore HLC on snapshot load and call _hlc.update() per node.
• Pass HLC through bulk_write and gossip.receive; reject stale gossip updates and skip missing-HLC overwrites on existing keys.
• Hold _lock during restore_from_file() and gc_expired() to prevent concurrent store corruption.

Validation

• Reproduced pre-fix: restored HLC was fresh (wall, 0, uuid) and successor write failed.
• Post-fix reproduction passes.
• pytest tests/test_enterprise_backup.py tests/test_gossip.py tests/test_rust_brain.py tests/test_rust_brain_concurrency.py — 37 passed.