Send RST_STREAM for a cancelled half-closed stream

[rp-arielrodriguez]

We found this with HTTP/2 multiplexing in HttpClient 5: when a GET timed out
on a shared connection and the server stayed silent, no RST_STREAM was sent.
The server kept working on the request, and the stream kept its slot.

H2Stream#abort() sets cancelled and calls requestOutput(), but the flag is
checked only in consumeHeader, consumeData and consumePromise. The output
loop skips local-closed streams, so a half-closed stream with a silent peer is
never reset.

Fix: H2Stream#resetIfCancelled() sends RST_STREAM(CANCEL) for a cancelled
half-closed stream. onOutput() calls it for each stream, outside the
connection window check, because RST_STREAM is not subject to flow control.

Backport: the main code applies cleanly to 5.4.x; the test file needs a small
merge.

Tests:

• New: testAbortAfterLocalEndStreamSendsRstStreamWithoutInboundFrames (fails
  on master, passes with the fix).
• New: testAbortAfterLocalEndStreamSendsRstStreamWithNoConnectionWindow (fails
  on master and on the previous revision of this PR, passes with the fix).
• ./mvnw -pl httpcore5-testing -am test: httpcore5 1960, httpcore5-h2 388,
  httpcore5-testing 443 (4 skipped), all pass. Checkstyle and RAT pass.

I used an AI assistant (Claude Code) for this change. I reviewed all of it and
I can answer questions about it.

[arturobernalg]

@rp-arielrodriguez RST_STREAM is not subject to flow control. Shouldn't a pending stream reset be processed regardless of connOutputWindow?

[rp-arielrodriguez]

@arturobernalg Agreed, thanks. I will apply the window check only to normal stream output, so a pending reset is sent even when connOutputWindow is 0, and add a test for it. Should the reset also skip the remote SETTINGS ACK check?

[ok2c]

@rp-arielrodriguez Cherry-picked to 5.4.x though I had to port the tests manually. Please double-check.