Use an atomic.Int64 as bodyWrapper read bytes counter

[dmathieu]

Closes #4895.

This switches the bodyWrapper struct to use an atomic.Int64 instead of a plain int64 integer to count read bytes.
This should solve the race condition mentioned in #4895, though unfortunately, this isn't being tested as I haven't been able to reproduce the issue within the codebase.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ee76330) 79.6% compared to head (cf65189) 79.6%.

Additional details and impacted files

@@          Coverage Diff          @@
##            main   #5080   +/-   ##
=====================================
  Coverage   79.6%   79.6%           
=====================================
  Files        152     152           
  Lines      10511   10511           
=====================================
  Hits        8371    8371           
  Misses      1944    1944           
  Partials     196     196           

Files	Coverage Δ
instrumentation/net/http/otelhttp/handler.go	87.3% <100.0%> (ø)
instrumentation/net/http/otelhttp/transport.go	96.9% <100.0%> (ø)
instrumentation/net/http/otelhttp/wrap.go	92.5% <100.0%> (ø)

[ash2k]

I think this fix may lead to incorrect metric reported, see my comment here #4895 (comment). This can be checked by having a reader for the body that is slow+a big body and a server that starts responding without consuming all of the incoming request's body.

[MrAlias]

I think this is still as improvement worth including. Slightly incorrect data is better then a panic.

Please do submit a change to address the data inconsistency you mention.

[ash2k]

I agree this is an improvement vs a data race. I'm not planning to work on this, but I've opened an issue so that this is not lost and someone can pick it up #5087.