Refactor CutSet.describe to enable parallel statistics computation

[pzelasko]

Resolves #1167. Below is a toy example showing how to use this with multiple cut sets (for simplicity, I'm replicating an existing one here).

from lhotse import CutSet
from lhotse.cut.describe import CutSetStatistics
from concurrent.futures import ProcessPoolExecutor


def work(cs): return CutSetStatistics().accumulate(cs)

if __name__ == "__main__":
    cuts = CutSet.from_file("libri-train-5.jsonl.gz")

    print("Sequential")
    cuts.repeat(100).describe()

    print("Parallel")
    with ProcessPoolExecutor(8) as ex:
        stats = list(ex.map(
            work,
            [cuts] * 100,
        ))
    stats = stats[0].combine(*stats[1:])
    stats.describe()