Copy a database with history via transaction logs. Based on code from Cognitect.

Continuous restore functionality that polls a TransactionStore and applies
new segments to a target database as they become available.

This namespace provides background restore that:
- Pre-fetches segments ahead of the consumer using core.async
- Uses exponential backoff on errors
- Supports graceful shutdown via a running? flag
- Logs structured maps for CloudWatch integration

LRU cache for long->long mappings with bounded size.

This is a simple LRU cache - it doesn't know anything about Datomic IDs
or entity indices. That logic belongs in the restore code that uses this cache.

A TransactionStore implementation that reads directly from a live Datomic database connection.

This store provides read-only access to transaction history from a source database,
presenting it through the same TransactionStore interface used by backup stores.
Useful for direct database-to-database cloning without intermediate storage.

(log/info {:msg "Segment restored"
           :segment-start 1000
           :segment-end 2000
           :duration-ms 1234})

JSON logging support for CloudWatch integration.

Log messages should be maps with a :msg key for the human-readable message.
Additional keys become structured data that CloudWatch Logs Insights can query.

Example:
```
(log/info {:msg "Segment restored"
           :segment-start 1000
           :segment-end 2000
           :duration-ms 1234})
```

clj -X:replicate \
  :source-client-config '{:server-type :cloud :region "us-east-1" :system "prod" :endpoint "https://..."}' \
  :source-db-name '"my-database"' \
  :target-client-config '{:server-type :cloud :region "us-west-2" :system "dr" :endpoint "https://..."}' \
  :target-db-name '"my-database-replica"'

CLI entry point for continuous live replication.

Run with:
```
clj -X:replicate \
  :source-client-config '{:server-type :cloud :region "us-east-1" :system "prod" :endpoint "https://..."}' \
  :source-db-name '"my-database"' \
  :target-client-config '{:server-type :cloud :region "us-west-2" :system "dr" :endpoint "https://..."}' \
  :target-db-name '"my-database-replica"'
```

AWS implementation of a TransactionStore on S3.

NOTES:

* Since s3 is eventually consistent it is possible that any reported data from
methods like `last-segment-info` are incorrect. This should be ok in most cases because backing up
a segment is an idempotent operation.
* S3 bills by request.
** Choosing a small number of transactions per segment leads to more PUT requests.
** The `last-segment-info` is the cheapest way to figure out where to resume, since
it is a single request. Using `saved-segment-info` lists all objects in the store, and can result
in a lot of requests and network overhead.