refactor: split large modules into focused submodules

Break up oversized single-file modules across nodedb-columnar,
nodedb-crdt, and the nodedb executor/engine layers:

- nodedb-columnar: compaction → compaction/{extract,segment,segments}
  mutation → mutation/{engine,flush,write}
- nodedb-crdt: state → state/{core,history,snapshot}
- nodedb executor: columnar_read → {bitemporal,convert,filter,scan,sort}
  columnar_write → {insert,read_prior,schema,spatial}
  timeseries/ingest → ingest + ingest_formats
- nodedb engine: timeseries/merge → merge/{cycle,o3,partitions}
  timeseries/partition_registry → {entry,lifecycle,persistence,query,rate,registry}
  crdt/tenant_state → {core,history,policy}
  timeseries: extract ilp_schema into its own file

Author: farhan-syah