PyTerrier Integration

[seanmacavaney]

This PR adds support for using rerankers in PyTerrier pipelines. It adds an optional [pyterrier] dependency and is constructed using a new .as_pyterrier_transformer() method (following the convention from LangChain).

The integration allows rerankers to be composed into larger retrieval experimental pipelines, e.g.,:

import pyterrier as pt
from rerankers import Reranker
from ir_measures import nDCG

dataset = pt.get_dataset('irds:msmarco-passage/trec-dl-2019/judged')
index = pt.Artifact.from_hf('macavaney/msmarco-passage.pisa') # load a PISA index for fast bm25 retrieval
reranker = Reranker('cross-encoder')
pipeline = index.bm25() % 100 >> dataset.text_loader() >> reranker.as_pyterrier_transformer()

pt.Experiment(
    [pipeline],
    dataset.get_topics(),
    dataset.get_qrels(),
    eval_metrics=[nDCG@10],
)

The pt_docs/index.rst is the documentation that will be pulled through to the PyTerrier documentation.

@Mandeep-Rathee (who co-authored it) and I have been using it for a while, and it's working great. We figure it would be helpful for others too.