feat: find endpoints asynchronously, cache errors

[rvagg]

Goals:

1. Make endpoint lookup parallel across all SPs
2. Don't error unless we don't have any endpoints—but log an error
3. Cache the error case too because it's not cheap and with a misbehaving SP it'll slow down every call for a particular piece

Critique of my design decisions welcome of course, there's a bunch of ways this could be done, I made some choices.

[codecov]

Codecov Report

Attention: 5 lines in your changes are missing coverage. Please review.

Comparison is base (2d540e3) 74.02% compared to head (dd7d1a7) 74.08%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #419      +/-   ##
==========================================
+ Coverage   74.02%   74.08%   +0.06%     
==========================================
  Files         149      150       +1     
  Lines        9823     9880      +57     
==========================================
+ Hits         7271     7320      +49     
- Misses       1802     1807       +5     
- Partials      750      753       +3     

Files	Coverage Δ
api/api.go	75.07% <100.00%> (+1.31%)	⬆️
retriever/endpointfinder/options.go	100.00% <100.00%> (ø)
retriever/endpointfinder/endpointfinder.go	93.67% <89.79%> (-0.78%)	⬇️

... and 2 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rvagg]

@gammazero PTAL I've switched to an options pattern for this but also slightly changed the parallelism such that it only forks a goroutine if there's something to look up, so the common call here will just bounce right through and collect values from the cache, no goroutine, no channels. Can you sanity check the logic for me please?

[gammazero]

LGTM with one question.