TCP exponential backoff fix

[tomDev5]

Currently, after a TCP retransmit, the socket timer was set to idle, which means that no actions except for keep-alive are required.
This is incorrect in the case where another TCP retransmission will be required.
This PR adds a test and fixes the problem.

And just to make sure I didn't miss anything - what if set_for_retransmit sets a timeout larger than the keep alive?
I assume that's OK because the socket won't close with unacknowledged data?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.83%. Comparing base (2ecd77f) to head (4917477).
Report is 6 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1020      +/-   ##
==========================================
+ Coverage   80.81%   80.83%   +0.02%     
==========================================
  Files          81       81              
  Lines       28385    28415      +30     
==========================================
+ Hits        22939    22969      +30     
  Misses       5446     5446              

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

[Dirbaio]

i'm not sure if restarting the retransmit timer here is 100% correct, because it's not yet guaranteed we're going to send a packet. Actually transmitting is done further below, which might fail because there's not enough buffers in the device for example.

i've opened #1023 which does this a slightly different way: it still does set_for_idle here so we only restart the retransmit timer if we did actually send a retransmission packet (because it's done after sending), and keeps the delay variable outside the timer (in the RTTE struct) so it doesn't get cleared.

i've included your test from this PR in it.