feat: Raft Snapshotting and Log Compaction

[Migorithm]

closes #180

Changes

Prior to this branch, the Raft log would grow infinitely. This branch adds the ability to periodically take a full snapshot of the state machine, save it to disk, and delete the old log entries to save space, as well as transfer those snapshots over the network to lagging nodes.

1. Configuration & Triggers (config.rs, multi_raft.rs)

• Introduced raft_snapshot_entry_threshold (default 10,000) with proper clap bounds validation.
• Added logic in the MultiRaftActor to periodically check if a shard group has applied enough
  entries to cross this threshold, triggering a snapshot creation.

2. Storage Layer & Compaction (impls/metadata_storage.rs, storage.rs)

• Reserved new RocksDB keys (0x03 for Snapshot Metadata, 0x04 for Snapshot Data, 0x05 for Applied
  Index).
• Implemented atomic snapshot persistence: A single WriteBatch now saves the new snapshot while
  issuing a Range Tombstone to delete the covered LogEntry prefix.
• Added the fail-fast assert_eq! check on startup to guarantee the surviving uncompacted log suffix
  is strictly contiguous.

3. Network Transfer & Validation (transient_state.rs, messages/rpc.rs)

• Added the InstallSnapshot and InstallSnapshotResponse RPC messages.
• Added chunk buffering logic in TransientState.
• Implemented strict bounds checking (the accepts() method we looked at) to ensure incoming network
  chunks are strictly sequential, don't exceed SNAPSHOT_CHUNK_BYTES, and match the declared
  size_bytes header.

4. The Staging State Machine (raft/state.rs, log_state.rs)

• Implemented a safe lifecycle for received snapshots: None -> Staged -> Active.
• handle_install_snapshot_request: Buffers chunks and stages them.
• publish_persisted_snapshot: The final step that flips the snapshot from Staged to Active
  (overwriting the live MetadataState and self.peers) after the actor confirms it was safely written
  to disk.

5. Integration Tests (it/raft/snapshot.rs)

• Added a robust new test suite specifically dedicated to testing snapshot creation, the chunked
  network installation process, and node recovery from disk snapshots.