Add support to data anonymizer for external mapping storage

[xyunxylona]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Enhance the reversible anonymization such that the mappings can be retrieved and saved to external stores like Redis.

Motivation

This is a concurrency/scalability request. We currently have a service running the data anonymizer, and it works fine when running on 1 instance, but when this service needs to scale out to 2 or more instances, different instances would have different mappings stored in memory.

E.g. anonymization request for John Smith arrives into instance 1. Instance 1 generates mapping <PERSON_1> <-> JOHN Smith. De-anonymization request for <PERSON_1> now arrives into instance 2. Instance 2 does not have the mappings, and cannot de-anonymize.

Proposal (If applicable)

Instead of storing the mappings in-memory by default, provide an option such that the mappings can be retrieved and saved to external stores like Redis. This also reduces the memory load of the service running the anonymizer as it does not need to keep all the mappings in memory. Additionally, this allows the mappings to be persisted across instance restarts.

[kuator]

It's been a year already, langchain devs, wtf