Make HTTP/2 response body close idempotent - #26

transportResponseBody.Close currently performs its reset logic again when called more than once. For an unfinished response, two calls emit two RST_STREAM(CANCEL) frames. After a response body has already been read to EOF, the first Close records the closed-body error and a second Close incorrectly emits a reset.

Treat a body already closed by the caller as a successful no-op. This preserves the first close's existing behavior while preventing repeated reset and flow-control side effects.

The regression test covers both cases:

• an unfinished response body emits exactly one RST_STREAM(CANCEL) across two Close calls
• a response body read to EOF emits no RST_STREAM across two Close calls

Test results

Passed on Go 1.24.1 (toolchain selected by the current upstream go.mod):

go test ./http2 -run '^TestTransportResponseBodyCloseIsIdempotent$' -count=1
ok  github.com/bogdanfinn/fhttp/http2  1.019s

go test -race ./http2 -run '^TestTransportResponseBodyCloseIsIdempotent$' -count=1
ok  github.com/bogdanfinn/fhttp/http2  2.043s

go test ./http2 -run '^TestTransportResponseBodyCloseIsIdempotent$' -count=10
ok  github.com/bogdanfinn/fhttp/http2  10.060s

The regression test fails against unpatched upstream as expected:

unfinished: received 2 RST_STREAM frames; want 1
after EOF: received 1 RST_STREAM frames; want 0

The full package command was also run:

go test ./http2 -count=1

It completed in 94.007s with one existing failure in TestTransportRejectsConnHeaders (Content-Length unexpectedly present). The same failure reproduces on a clean checkout of the upstream base commit, so it is unrelated to this change.