Delete old BBR and pacer implementations in the recovery/congestion directory.

[antoniovicente]

There are big known problems in these implementations.

We recommend use of:

• the cubic implementation, without pacing
• the bbr2 implementation from the gcongestion directory, with pacing. Note that pacing only works correctly on Linux and if TxTime is available (qdisc fq kernel module enabled).

[reneleonhardt]

* the bbr2 implementation from the gcongestion directory, with pacing.

But this PR deletes brr and brr2? What about switching to brr3?

Other users report much better experiences with brr than cubic, what are those big known problems?

[antoniovicente]

* the bbr2 implementation from the gcongestion directory, with pacing.

But this PR deletes brr and brr2? What about switching to brr3?

Other users report much better experiences with brr than cubic, what are those big known problems?

The main motivation behind this PR is to delete the pacer in the congestion directory. Since BBR implementations in that directory have a hard dependency on it, I think it is time for them to also be deleted.

All of our investments are on making the bbr2 implementation in the gcongestion directory work well even in cases where Linux TxTime is not available. I have several planned changes in the near future to fix some recently found bugs and make this implementation closer to the BBR RFC draft.

The config option to select "bbr2" still works. I can also keep the "bbr" option, but point it to the bbr2 implementation in the gcongestion directory.

[LPardue]

LGTM modulo nits