Fix bugs in DCTCP on TCP events

[rscheff]

Summary

DCTCP reactions to after_idle and post_recovery would disable
themselves after the first call.
Also, have consistent fall-through to NewReno for non-ECN
sessions.

[rscheff]

Comment from Midori (original author of this code):

Thank you for your message and improving dctcp implementation. The description of two bugfix request makes sense. I have two comments. First, remove white space at line126 in the D20550 patch before submission. Second, you
should remove a change at line353 in the D20550 patch. If I understand correctly, it does not affect performance. Please point out if there is my misunderstanding.

[rscheff]

sys/netinet/cc/cc_dctcp.c

127 ↗

(On Diff #58358)

remove white space at line126 in the D20550 patch before submission

353 ↗

(On Diff #58358)

MidorI: remove a change at line353 in the D20550 patch. If I understand correctly, it does not affect performance

R: If other parts of the code decided to send out an ACK right away (e.g. leaving zero-window episodes), we should not clear the ACKNOW flag here.

[rscheff]

• - missed rename of variable

[rscheff]

sys/netinet/cc/cc_dctcp.c

127 ↗

(On Diff #58358)

Leaving whitespace, as that separates different ack types.

using sacked data (CCV(ccv, sackhint.delivered_data) here depends on the RFC6675 chain of patched (D18892, D18985).

[tuexen]

sys/netinet/cc/cc_dctcp.c

293 ↗

(On Diff #58729)

This should read

newreno_cc_algo.cong_signal(ccv, type);