THRIFT-6023: Add HTTP transport support to PHP cross-tests

[sveneld]

Adds HTTP transport support to the PHP entry of the cross-test matrix. PHP was the only major-language entry stuck on buffered/framed while Python, Go, Node, C++, Java, Lua, Dart, JS, D, and Haskell already declare http and verify HTTP interop.

Server — TestServer.php detects --transport=http in CLI mode and pcntl_exec()s into PHP's built-in web server (php -S 127.0.0.1:$port TestServer.php). The same script re-enters under SAPI cli-server, reads the request via TPhpStream from php://input, dispatches it through the existing processor, and writes the response to php://output. The original -d flags from the launcher cmdline (e.g. -dextension=thrift_protocol.so) are preserved by parsing /proc/self/cmdline, so the accel extension and sockets stay loaded under the spawned server too. The cross-runner's socket.connect_ex readiness probe works because php -S TCP-binds the port immediately.

Client — TestClient.php gains an if ($MODE == 'http') branch that uses TPsrHttpClient (added in THRIFT-6010) against http://127.0.0.1:$port/. TPsrHttpClient buffers internally, so no TBufferedTransport/TFramedTransport wrapping.

tests.json — adds "http" to PHP server and client transports; bumps client timeout from 6s → 10s to absorb HTTP overhead.

composer.json — adds guzzlehttp/guzzle: ^7.8 to require-dev so the PSR-18 auto-discovery in TPsrHttpClient resolves a concrete HTTP client at cross-test time.

JIRA: THRIFT-6023. Draft until cross-test CI validates the new php↔php / php↔py3 / php↔go HTTP matrix cells.

• Did you create an Apache Jira ticket? (Request account here, not required for trivial changes)
• If a ticket exists: Does your pull request title follow the pattern "THRIFT-NNNN: describe my issue"?
• Did you squash your changes to a single commit? (not required, but preferred)
• Did you do your best to avoid breaking changes? If one was needed, did you label the Jira ticket with "Breaking-Change"?
• If your change does not involve any code, include [skip ci] anywhere in the commit message to free up build resources.

[kpumuk]

The last commit should hopefully be unnecessary with #3514.

For Java, the issue is that it is the only cross-test that actually verifies the one-way semantics, as in that HTTP server must return the response immediately, and then run the handler afterwards (so that the client is as fast as possible).

[sveneld]

Thanks, i will update PR

[sveneld]

@kpumuk Thanks — addressed both points in 025863b:

• Implemented oneway fast-response in the cli-server router: the request is peeked for TMessageType::ONEWAY, and on a match we send an empty HTTP 200 (Content-Length: 0) and pcntl_fork() the handler so the client's oneway call returns immediately, matching the THttpServer pattern from THRIFT-6021. This drops the four php-java_*_http-ip known_failures entries.
• Removed the four php-cpp_*_http-ip known_failures entries on the assumption that THRIFT-6021: Make C++ HTTP client work with oneway requests #3514 lands before this PR — let me know if you'd prefer I re-add them as a temporary hedge until THRIFT-6021: Make C++ HTTP client work with oneway requests #3514 merges.

While I was in there also collapsed the if/elseif cascade in TestClient.php to two match expressions and dropped the dead TSocket assignment / unused $hosts/$logger.