Chunked request with trailers incorrectly closes connection instead of keeping alive

[wi-adam]

Bug Description

When an HTTP/1.1 chunked request body includes trailers, the connection is closed instead of being kept alive for reuse.

Root Cause

In src/proto/h1/conn.rs, poll_read_body transitions to Reading::Closed when a trailers frame is received (line 396). It should transition to Reading::KeepAlive, matching the behavior for a completed data frame at EOF (line 378). Both cases indicate body completion, but only the data path allows connection reuse.

This causes try_keep_alive() to close the connection when it sees Reading::Closed, preventing any subsequent request on the same connection.

Reproduction

Send a chunked POST with trailers on a keep-alive connection, then attempt to send a second request.

client > server
  POST / HTTP/1.1
  trailer: chunky-trailer
  host: example.domain
  transfer-encoding: chunked
  
  5
  hello
  0
  chunky-trailer: header data

server > client
  HTTP/1.1 200 OK
  content-length: 2
  date: Thu, 26 Mar 2026 17:30:47 GMT
  
  ok

client > server
  GET /quux HTTP/1.1          ← second request, reusing connection
  Host: example.domain
  Connection: close

server > client
  [FIN] connection closed     ← bug: server closed instead of responding

Without trailers, the same sequence works correctly and the second request receives a response.

Fix

PR #4043