Allowing users to submit Dataset/Benchmark/Evaluation/Transcripts

[tianyi-ant]

Ideally we want to move towards a zoo of various benchmarks with cost/latency/performance tradeoffs. This should be relatively high priority.