fix(http): time out stalled transactions without killing switched protocols

[nickanderson]

A server that accepts a request and then never answers left the response read
outstanding indefinitely, wedging the client silently — alive, no crash, no log
output, recoverable only by restart. Reproduced in a VM on both 5.0.5 and 5.1.0.

The gap was ReadHeader(): AsyncReadNextBodyPart already had a 5 minute timer,
but nothing bounded the wait for response headers.

This arms Beast's stream timeout before each operation of a transaction and
disarms it at the 101 Switching Protocols handover. That handover is what the
previous attempt (d8ddf6da / 4aa01ee8) got wrong: it left the timeout armed,
and Beast's timeout_handler calls close() on expiry regardless of who owns the
socket by then, which killed mender-connect's forwarded WebSocket after five
minutes and got the fix reverted under MEN-9433.

Connect is intentionally not covered — asio::async_connect runs on
lowest_layer(), the raw socket, which bypasses the stream timeout entirely.

Timeout is ClientConfig::stream_timeout_seconds, default 300s. No client
endpoint long-polls, and artifact download is a per-chunk body read (unchanged).

Tests

Both are red/green verified — each fails when its own half of the fix is removed:

test	without the fix
HttpTest.TestStalledResponseTimesOut	hangs to the event-loop watchdog (5006 ms)
HttpForwarderTests.ProtocolSwitchSurvivesIdlePastStreamTimeout	fails at 1219 ms — MEN-9433 exactly

The forwarder already had five ProtocolSwitch* tests, but all exercise immediate
data flow; none left the connection idle, which is why MEN-9433 escaped.

Full suites pass: http_test 47/47, http_forwarder_test 13/13.

End-to-end on 5.0.5 with a stalled deployments/next and the reporting customer's
state scripts: unpatched logs nothing for 420 s, patched logs
Error while polling for deployment: The operation timed out at exactly t+300 s,
runs the SyncError/IdleEnter scripts, and resumes polling.

Ticket: ME-731

[mender-test-bot]

@nickanderson, start a full client pipeline with:

• mentioning me and start client pipeline

---

my commands and options

You can prevent me from automatically starting CI pipelines:

• if your pull request title starts with "[NoCI] ..."

You can trigger a client pipeline on multiple prs with:

• mentioning me and start client pipeline --pr mender/127 --pr mender-connect/255

You can trigger a client pipeline for a specific Mender Client release with:

• mentioning me and start client pipeline --release 6.0.x (can be given multiple times)
• by default, a pipeline is triggered for each supported release the component is a part of

You can trigger GitHub->GitLab branch sync with:

• mentioning me and sync

You can print PR statistics for a repository with:

• mentioning me and print fast pr stats (Team stats only)
• mentioning me and print full pr stats (Detailed report)
• options: --repo <repo>, --team <name>, --all-repos, --exclude-drafts, --exclude-user <user>
• mentioning me and print full pr stats --repo mender --all-repos --exclude-drafts

You can deploy a review app with:

• mentioning me and start review app (OS environment)
• mentioning me and start review app enterprise (Enterprise environment)

You can run e2e tests against a deployed review app with:

• mentioning me and start review tests (defaults to os environment)
• mentioning me and start review tests enterprise (for enterprise environment)

You can cherry pick to a given branch or branches with:

• mentioning me and:

 cherry-pick to:
 * 1.0.x
 * 2.0.x

[mender-test-bot]

Merging these commits will result in the following changelog entries:

Changelogs

mender (ME-731/master)

New changes in mender since master:

Bug Fixes

• Fixed the client hanging indefinitely when a server accepts a request but
  never sends a response, which left the device unmanageable until it was restarted.
  (ME-731, MEN-9433)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!