chore: Mmr Checkpoint

[sergerad]

Warning

Blocked until incremental serialization of Mmr enabled by miden-vm (0xMiden/miden-vm#3585). Current impl performs full mmr file write every time.

Summary

Relates to #1697.

Why:

• The chain MMR had no persisted form: it was rebuilt from every block header commitment in
  SQLite on every startup, the only remaining load-time cost that grows with chain height.
• Operators get startup time that no longer scales with chain length; no functional or API
  change.

How:

• Added ChainMmrCheckpoint: an on-disk cache of the derived MMR (<data-dir>/chainmmr.bin,
  version-prefixed serialized Blockchain).
• load_mmr restores the checkpoint and tops it up with only the block commitments committed
  since it was taken (new select_block_header_commitments_from query), then verifies the
  result against the latest header's chain commitment — the same check the rebuild path uses.
  A missing, corrupt, or divergent checkpoint transparently falls back to a full rebuild.
• The write worker refreshes the checkpoint on graceful shutdown, so the next startup only
  tops up blocks committed after that point.

Changelog

[[entry]]
scope       = "node"
impact      = "changed"
description = "Store startup now restores the chain MMR from an on-disk checkpoint and tops it up with only the blocks committed since, instead of rebuilding it from the database on every boot."