test: drip server closes the mid-response streaming coverage gaps

[puzza007]

The local httpbin stack buffers whole responses behind its proxy, so two streaming behaviors have been test-blind since the streaming work landed — reachable only in the TLA+ model, never against a real socket: a terminal error arriving after chunks were already delivered, and a cancel landing between headers and done. Both were disclosed as accepted gaps in #182.

katipo_drip_server closes them: a ~90-line single-connection HTTP/1.1 responder that can advertise a larger Content-Length than it sends (closing early yields a deterministic partial_file mid-body), deliver timed body pieces so a response is genuinely in progress when the interesting event happens, stall with the socket open, and report peer-close to the owning test. It's linked to the test process, so failures tear it down.

Three new cases in the async group. streaming_error_mid_body: headers, at least one chunk, then the terminal partial_file — the error-after-chunks sequencing previously only model-verified. streaming_timeout_mid_body: headers and a chunk, then operation_timedout while the server stalls. And streaming_cancel_mid_stream, the strongest of the set: after a cancel between headers and done, in-flight chunks may drain but no terminal message ever arrives — and the stalled server's recv observes the connection close, the end-to-end proof that cancel_conn in the C port actually aborts the transfer rather than merely having its late output dropped. That C path had never been exercised against a live socket.

The suite also gains expect_stream_headers/expect_first_chunk helpers shared with the previously inline receives in streaming_flow_control. 202 tests, dialyzer, xref, and lint pass.

Top of the stack: #180 → #182 → #183 → #184 → #185 → this. Merge bottom-up; I'll rebase each layer as the one below lands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Vf2gxZLpxT4HvBfZKCgP3g

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.52%. Comparing base (bec37ad) to head (e14096b).

Additional details and impacted files

@@                      Coverage Diff                      @@
##           bounded-admission-timeout     #186      +/-   ##
=============================================================
+ Coverage                      78.23%   78.52%   +0.29%     
=============================================================
  Files                              9        9              
  Lines                           1369     1369              
=============================================================
+ Hits                            1071     1075       +4     
+ Misses                           298      294       -4     

Flag	Coverage Δ
c	78.52% <ø> (+0.29%)	⬆️
erlang	78.52% <ø> (+0.29%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.