net: add some more practical tests for TCP keep-alive mechanism

[panjf2000]

This is a tracking issue of testing the TCP keep-alive mechanism in Go.

The current test suites are based on this previous discussion. They were good enough back then, but they might miss something like #65809, which directs us here, to investigate the possibility of introducing some additional tests that will reflect the real world.

CC @bcmills @ianlancetaylor @neild

[garrickisgross]

I added a dragonfly regression test for #65809 just in case someone decides to switch back to milliseconds or something.

Additionally, would the Go team be interested in some kind of config test here in tcpconn_keepalive_test.go that does something like the below?

• sets keepalive config A, reads back
• sets keepalive config B, reads back
• disables keepalive, reads back
• asserts transitions are reflected in kernel state

If so, I can add it before I submit.