[ENHANCEMENT] add the ability to filter spans by document evaluations

[axiomofjoy]

Users debugging RAG want to filter their spans to know whether the fault lies with the retriever or the LLM. We need a way for the user to filter document evaluations using our query DSL so they can write filters such as:

document_evals["relevance"].label == "relevant" and evals["qa_correctness"].label == "incorrect"

https://arize-ai.slack.com/archives/C04R3GXC8HK/p1722442515198979?thread_ts=1722436956.324489&cid=C04R3GXC8HK

[RogerHYang]

Because evals for docs and spans are on different spans, this is essentially blocked, unless we can apply a filter for evals on the root span and still include sub-spans (i.e. where the doc evals are) in the calculation of document evaluation summaries.