Test HTTP/2 benchmarks with artificially induced latency

[halter73]

This might be interesting for all of our end-to-end benchmarks, but benchmarking with high-latency connections is particularly important for HTTP/2 because Kestrel and HttpClient implement their own flow control logic. This came up during discussion with some msquic folks about the relatively small 128 KB default size of Kestrel's HTTP/2 connection window.

An artificial latency inducer is probably most easily done with Kestrel Connection Middleware.

@nibanks