Match chunked as the final transfer coding, order-independently

[yhirose]

Based on #2487 by @metsw24-max — reworked implementation, so opening separately to make the diff reviewable.

Problem

is_chunked_transfer_encoding compared the whole Transfer-Encoding field value against "chunked", so a message whose final coding is chunked but which names another coding first ("gzip, chunked", valid under RFC 9112 §6.1 where chunked must be the final coding) was read as unframed. On a keep-alive server the body is then left in the socket and parsed as the next request (smuggling). open_stream repeated the check case-sensitively with a raw ==, desyncing the client stream the same way.

Change

• Match the last coding token of the Transfer-Encoding value case-insensitively instead of comparing the whole field, and route open_stream through the same helper so server and streaming client agree.
• Scan all Transfer-Encoding lines rather than only index 0. Headers is an unordered_multimap whose duplicate-key iteration order is not portable (this is what broke CI on the original PR: libstdc++ and libc++ disagree), so the final coding of a multi-line field cannot be determined reliably. In that ambiguous case we fail safe — any multi-line field that names chunked is treated as chunked. A mis-parse only closes the connection, whereas the opposite error enables smuggling.

Tests

ChunkedTransferEncodingTest covers single-coding (case-insensitive), chunked as the final coding of a list, non-final chunked, absent/empty, and order-independent multi-line cases. Builds and passes on both the header-only and split (test_split) builds.

Difference from #2487

#2487 read only the first Transfer-Encoding line via get_header_value(..., 0) and dropped its multi-line test because of the unordered_multimap ordering issue. This version handles the multiple-line case deterministically by failing safe, so no ordering assumption is needed.

🤖 Generated with Claude Code

[yhirose]

@metsw24-max thanks for sending #2487. This PR is the revised version that I made with Opus. I think this is a more comprehensive solution. Please let me know if you have any question.