instructlab-training initial release take 2

This introduces the instructlab library as a package in the instructlab package namespace.

To install it:
pip install instructlab-training

And to install it with flash-attn and other CUDA-dependent packages, you can use
pip install instructlab-training[cuda]

Here's how to use it:

from instructlab.training.config import TorchrunArgs, TrainingArgs, run_training

torchrun_args = TorchrunArgs(
    nproc_per_node = 1,  # 1 GPU
    nnodes = 1,  # only 1 overall machine in the system
    node_rank = 0,  # rank of the current machine
    rdzv_id = 123,  # what ID other nodes will join on
    rdzv_endpoint = '0.0.0.0:12345'  # address where other nodes will join
)

training_args = TrainingArgs(
    # specify training args here
)


run_training(torch_args = torchrun_args, train_args = training_args)