perf: Make writing of ClojureCat ensembles more memory efficient

[Schaechtle]

What does this do?

This uses (prn) to print a ClojureCat ensemble and re-directs the output in dvc.yaml to a file.

Why do we want this?

Solves previous memory issues when running the dvc pipeline, e.g. with large CES models.

Why is this a draft?

This is a draft because resulting ensemble for CES is too large to be loaded into iql.query:

 clj -M -m inferenceql.query.main --model m=data/xcat/xcat.merged.edn --lang permissive
Execution error (OutOfMemoryError) at jdk.internal.util.ArraysSupport/hugeLength (ArraysSupport.java:649).
Required array length 2147483639 + 9 is too large

Unless, there is an obvious fix to the above, I guess we still merge?

@zane comments welcome.

[zane]

Combining the slurp with the EDN reading will further reduce memory usage.

[Schaechtle]

Update: after chatting about this with Zane, we decided that we should use iql.query's slurp/model and fix the memory issues there.