fix(server): stop writing a second response after a failed upgrade

[geovannewashington]

Summary

 HandleConnectionV1 / HandleConnectionV2 / revdial.ConnHandler
   conn, err := upgrader.Upgrade(...)      # on error, gorilla already wrote a 4xx
   if err != nil
-    write 500 with err.Error()            # ignored status, error text appended to the body
+    log the cause at Warn                 # with request-id, tenant, uid (revdial: dialer logger, uuid)
+    return nil                            # an error would make Echo write again

Revdial now resolves its dialer before the upgrade so it can log through it. A map entry that is not a *Dialer gets a 400 instead of an upgraded connection closed at once. Only *Dialer values are stored, so no caller reaches that path.

Closes shellhub-io/team#346

Evidence

• Before: a GET without upgrade headers to /ssh/connection, /agent/connection or /ssh/revdial:

  400
  Bad Request
  websocket: the client is not using the websocket protocol: 'upgrade' token not found in 'Connection' header
  

  The server logs http: superfluous response.WriteHeader call instead of the cause.

  After: new cases in TestHandleConnectionV1StatusCodes, TestHandleConnectionV2StatusCodes and TestConnHandlerAnswersANonUpgradePickupWithOnlyTheUpgradersResponse expect status 400 and body Bad Request\n. They fail when the second write is restored and pass with this change. The full server module test suite passes.

Merge Danger

Door: two-way

A real agent sends a valid upgrade request, so its handshake does not change. No migration or wire format changes.

Blast Radius: tunnels

Only failed upgrades on the agent connection and revdial pickup paths behave differently: no second write, and a Warn log line. Any client that sends a plain GET to those paths now adds a Warn line to the logs.

[geovannewashington]

/review

[github-actions]

Claude finished @geovannewashington's task in 2m 43s —— View job

---

Code Review Complete

Reviewed 4 files on all three axes: Standards (repo rules, language conventions, tests, code smells), Spec (does it implement shellhub-io/team#346?) and Correctness (security and runtime defects). No new issues found. The code looks good as-is.

The earlier judgement call about the duplicated unknown dialer guard is fixed. One type assertion on the Load result now handles both a missing key and a non-*Dialer entry. The expanded doc comments on HandleConnectionV1, HandleConnectionV2 and ConnHandler match what each function returns. dialer.logger cannot be nil at the new log site, because the only dialers.Store call stores a *Dialer built by NewDialer. Nothing in cloud/ uses revdial or the changed handlers.