Review read failures impacting writes

[sbordet]

Jetty version(s)
12

Description
Read failures should not impact writes (too much), as a write might still be possible even if the read failed.
Typical examples are early EOF in reads -- they should not impact writes.

The idea is that if a read failure is catastrophic, we would attempt the write anyway and will get the failure through the write, rather than assuming that the writes will fail.

Child issues / prs

• Fix buffer leak in HttpClientStreamTest.testUploadWithConcurrentServerCloseClosesStream() #10431
• Fix buffer leaks in FCGI and H3 HttpClientIdleTimeoutTest #10432

[sbordet]

I think this should also help with #10234, because idle timeouts should be treated as read failures and always as hard failures for pending writes.

[lorban]

While investigating a server buffer leak in HttpClientStreamTest.testUploadWithConcurrentServerCloseClosesStream() we discovered that stopping a connector while there is an in-flight request does not notify the failure listeners.

While not exactly the same issue, this should be taken into account too.

[sbordet]

FIxed by #10948.