use SO_RCVBUFFORCE to force receive buffer increase on Linux

[MarcoPolo]

Closes #3795

[marten-seemann]

There's no downside to always calling SO_RCVBUFFORCE, is there?

If that's the case, you could just do it as part of setReceiveBuffer, no?

[MarcoPolo]

A couple a reasons to keep it here:

1. unix.SO_RCVBUFFORCE is only defined for GOOS=linux. So this would fail if we put it in setReceiveBuffer as is, since that's shared.
2. Having it as a separate function lets us test only this thing. However, I'm not sure this will actually work since CI runner might not have CAP_NET_ADMIN (I don't when running this as a normal user).

[codecov]

Codecov Report

Merging #3804 (baca871) into master (6b74a9a) will increase coverage by 0.17%.
The diff coverage is 100.00%.

❗ Current head baca871 differs from pull request most recent head 9b08488. Consider uploading reports for the commit 9b08488 to get more accurate results

@@            Coverage Diff             @@
##           master    #3804      +/-   ##
==========================================
+ Coverage   83.97%   84.14%   +0.17%     
==========================================
  Files         142      141       -1     
  Lines       14208    14000     -208     
==========================================
- Hits        11930    11779     -151     
+ Misses       1846     1795      -51     
+ Partials      432      426       -6     

Impacted Files	Coverage Δ
sys_conn_helper_nonlinux.go	100.00% <100.00%> (ø)
transport.go	64.07% <100.00%> (-1.59%)	⬇️

... and 2 files with indirect coverage changes

[marten-seemann]

Thank you for working on this @MarcoPolo!

One minor change was still needed: We need to ignore the error from SetReadBuffer. Other than that, I modified the CI setup a bit, so that we can test this (it's even included in Codecov, which is pretty cool).

[marten-seemann]

Merging this now, since I need to build on top of it to increase the send buffer as well (#3811), and this is blocking progress on the GSO PR.