CutSet.from_files constructor for random order multi-file cutsets

[pzelasko]

An example of typical usage is:

# We start with a big cut set that refers to all of the data.
big_cutset = CutSet.from_file("100k_hour_cuts.jsonl.gz")

# It's being split into smaller parts below.
big_cutset.split_lazy("shards", chunk_size=1000)

# Now we can iterate over the shards, randomizing their order on each iteration.
# Randomization is enabled by default.
sharded_cutset = CutSet.from_files(Path("shards").glob("*.jsonl.gz"))

# Order of shards will be randomized
for cut in sharded_cutset:
    pass

# This time the order will be different
for cut in sharded_cutset:
   pass

# We can randomize the cuts across the shards as well in a streaming fashion
# (note: this happens automatically in samplers when shuffle=True)
cuts = sharded_cutset.shuffle(buffer_size=50000)

# Works as usual with samplers
sampler = DynamicBucketingSampler(sharded_cutset, ...)

[pzelasko]

I added one more improvement: by default we'll use Python RNG here so it's automatically different on each script instantiation. The user can specify the seed if they want reproducibility.