fix: measure "connect time" per-peer for peer scoring

[rvagg]

Prior to this, the "start" is the same for all peers, regardless of whether they come in from the indexer at different times; this makes it strictly local to the peer so we can score it accordingly.

I don't think this should make a material difference if we're getting the graphsync and http peers in an initial batch from the indexer. The current start-time calculation begins from before we even get candidates back, but they are all the same start. The actual value for this only matters for scoring candidates so mostly it's the relative value that matters. What we intended for this, however, was that if this number is 0 or close to it, it means we're already connected to the peer so we should try them first.

TransportProtocol is a public interface so this would technically be a breaking change.

[codecov-commenter]

Codecov Report

Merging #431 (4caba91) into main (a7c7002) will decrease coverage by 0.08%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #431      +/-   ##
==========================================
- Coverage   76.65%   76.57%   -0.08%     
==========================================
  Files          85       85              
  Lines        6377     6377              
==========================================
- Hits         4888     4883       -5     
- Misses       1238     1241       +3     
- Partials      251      253       +2     

Files Changed	Coverage Δ
pkg/retriever/graphsyncretriever.go	86.41% <100.00%> (+0.08%)	⬆️
pkg/retriever/httpretriever.go	86.82% <100.00%> (ø)
pkg/retriever/parallelpeerretriever.go	92.85% <100.00%> (-0.04%)	⬇️

... and 5 files with indirect coverage changes