Fix Spark TTFB calculation to exclude Graphsync retrievals

[pyropy]

Descrption

Current Spark TTFB score include both values from HTTP and Graphsync retrievals. This is not exactly correct behaviour as TTFB reported for Graphsync is the time when the whole retrieval has been performed by lassie and on the other hand time reported for HTTP is the time when the first byte has been received.

We want to exclude Graphsync retrievals from Spark TTFB score by using only retrievals done over HTTP.

Context from Slack conversations we previously had:

@juliangruber and I have been reviewing the Spark TTFB documentation I’ve been drafting during our 1:1s. We realized it would be best to exclude Graphsync measurements from the TTFB calculation, as they do not accurately reflect the actual TTFB of SPs.

[bajtos]

OTOH, by including Graphsync in the TTFB, we are creating an incentive for the community to push SPs to abandon Graphsync and start using HTTP.

How much work would it require to add HTTP-TTFB (similarly to how we added HTTP-RSR)?