perf(records): add in-process caches and benchmark tooling

[zzacharo]

Cache repeated DB lookups in the hot path of the records stream:

• contributors.py: cache affiliation mapping lookups
• parent.py: cache submitter email → user_id across records
• request.py: cache reviewer name/email → user_id across records
• log.py: batch CSV flushes every 100 rows instead of per-row

[zzacharo]

Also, potential further improvements:

• checking fix(performance): replace expensive deepcopy() calls with lighter alternatives inveniosoftware/invenio-rdm-records#2367 (we do a lot of deepcopy calls inside the migrator-kit)