address compliance issues

[sleepybishop]

This pr addresses the findings from h2o/h2o#3607.

The picotls related items are deferred until h2o/picotls#617 is merged.

[kazuho]

Thank you very much for the PR! I'll take a look and merge this.

[kazuho]

Excuse me, based on my understanding that this PR only to change the two points listed below, I would close without doing anything:

Persistent congestion: it is not required by QUIC. RFC 9002 is merely "exemplary" (Abstract of RFC 9000), and CUBIC relies on QUIC PTO to handle the same case. Therefore, there is little need, if any, to implement it.

Use of unified loss detection rather than per-epoch states. This also is not a conformance issue, since loss detection is specified by RFC 9002. This is an intentional design choice and I do not see a compelling reason to adopt a more complex design.

I will post the same message to the original issue once I finish going through all of them.

[sleepybishop]

Fair, thanks for taking a look.

[kazuho]

Thank you so much for all your contributions. I highly appreciate them.