Handle all error cases of downloading and storing persistent states with parts

[SmaGMan]

Scope

• Update cold boot logic in cold_boot.rs to persist the metadata file in temporary storage immediately after downloading the main state part, and ensure it is removed when temporary persistent state files are cleaned up.
• If a temporary persistent state file already exists, compare its size with the size from the FoundState response; if it differs, delete the temporary file and re-download.
• Before downloading the persistent state file, check for an existing state in DB:
  • If the state exists but its split depth does not match FoundState, return an error.
  • If the state exists and its split depth matches FoundState, reuse parts info from the shard state handle.
• Download only parts that are not yet stored; check the presence of each part before requesting it.
• Add context to every error propagation so logs clearly identify where the failure occurred.