[zyuhao7]

Summary

This combines and supersedes #2652 and #2653, rebased onto current master.

Content-Length parsing is validated in one shared helper (parseContentLength). A value that is not 1*DIGIT — empty, signed, containing whitespace, with trailing junk, or overflowing — is now a framing error instead of being silently reinterpreted. Repeated values must agree, and Content-Length with Transfer-Encoding is rejected. All of these let two peers disagree about where a body ends, which desynchronizes keep-alive and pipelined connections.

Request side (RFC 9110 §8.6, RFC 9112 §5.1, §6.3)

• Reject a Content-Length that is not 1*DIGIT.
• Reject a repeated Content-Length whose value differs from the first.
• Reject whitespace between a field name and its colon.
• Reject a malformed header line instead of treating it as end-of-headers.

Response side (RFC 9112 §6.3, §7.1)

• Reject a malformed or repeated Content-Length.
• Reject Content-Length together with Transfer-Encoding.
• Recognize the final transfer coding case-insensitively and accept parameters on it.
• Consume the whole chunked trailer section instead of a single line.
• Treat 204, 304 and informational responses as bodyless.
• Never read a body for a response to HEAD.
• Skip 1xx responses in the HTTP client and keep waiting for the final response.

Tests

• New response-parser unit tests cover malformed/overflowing/repeated Content-Length, CL+TE, case-insensitive chunked coding with parameters, trailers, bodyless statuses, informational responses and HEAD.
• New raw-socket integration tests exercise the request framing rules end to end (ContentLengthFramingTest.cc).
• unittest: 1986 assertions in 112 test cases, all passing.
• Framing integration tests: 6/6 passing against integration_test_server.

Fixes #2648