memoize dataset length for eval sample packing

[bursteratom]

Description

Fix for issue#1966, where eval_sample_packing=True caused evaluation being stuck on multi-gpu.

Motivation and Context

In issue#1966, evaluation on sample packed dataset on multiple GPU got stuck on RANK 0 gpu, when calling __len__(self) in MultipackBatchSampler. This is due to a change in transoformers' handling on callbacks during evaluation loop. This PR modifies MultipackBatchSampler.__len__(self) in src/axolotl/utils/samper/multipack.py such that it gets the length of eval_dataloader across all ranks.

How has this been tested?

Tested on example/llama-3/qlora-1b.yml

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)