fix(http): lock HttpServer start/restart via start_locked (Snapmaker #870 residual)

[aceRage]

Description

Surgical extract of the remaining Snapmaker/OrcaSlicer #870 lifetime/race work onto EdgeSlicer main. This does not re-land the MQTT stack already present from Snapmaker Snapmaker#847/Snapmaker#855 via Edge #22 / #28.

HttpServer::start() still created the IO server without holding m_server_mtx, and restart() called start() (including start_health_check()) while holding that lock. That raced is_healthy() on the health-check thread (UAF / double restart) and could deadlock a join of a retired health-check thread blocked in is_healthy().

Residual vs Snapmaker Snapmaker#870 (6faa813a3c)

Marker	Edge main (pre this PR)	This PR
MqttClient::create() + self_ weak_ptr	Present (#28)	no change — MQTT.cpp/.hpp untouched
tearing_down_ + cb_mtx_ / topics_mtx_	Present (#22/#28)	no change
Moonraker_Mqtt::m_client_mtx + get_mqtt_client(_tls) snapshots	Present (#22)	no change
m_cbs_mtx	Present (#22)	no change
SSWCP_MachineConnect_Instance::release_pin_client + pin CallAfter	Present (#28)	no change
HttpServer m_server_mtx around stop() / restart() / is_healthy()	Present (#22)	kept
HttpServer::start_locked()	Missing	ported
start() / restart() hold lock across destroy+recreate; health check after unlock	Missing (start() unlocked create; restart() called start() under the lock)	ported
start_http_server atomic (io-thread write vs unlocked readers)	Missing	ported (same HttpServer files as start_locked)
SSWCP sw_mqtt_connect failure callback captures engine as weak_ptr	Missing (shared_ptr self-pin)	ported (callback body kept as Edge Disconnect; no Snapmaker SNAP_LOG_BATCH / session-id)

Files

• src/slic3r/GUI/HttpServer.hpp — declare start_locked(); std::atomic<bool> start_http_server
• src/slic3r/GUI/HttpServer.cpp — extract body; lock in start() / restart(); start_health_check() after unlock
• src/slic3r/GUI/SSWCP.cpp — weak_ptr engine capture in sw_mqtt_connect only

Not touched: Ultra HttpServer login handlers (ResponseLoginFailed / BBL_LOGIN_LOCALHOST_URL), SSWCP Ultra unload/archive customs, RemoteSnapmaker.cpp, MQTT.cpp/.hpp, MoonRaker.cpp/.hpp.

Intentionally not ported (re-diff, not weak_ptr/CallAfter residuals)

• Moonraker time_sync_manager_snapshot() / m_time_sync_manager_mtx — mutex snapshot of an already-present weak_ptr<TimeSyncManager> capture, not a missing CallAfter/weak_ptr guard. Edge still copies time_sync_manager_ without that mutex.
• Moonraker disconnect(): Fix MQTT disconnect crash: lifetime & race fixes in MQTT/MoonRaker/HttpServer Snapmaker/OrcaSlicer#870 clears the message callback before resetting time_sync_manager_; Edge still resets first then SetMessageCallback(nullptr). Left as-is to avoid a MoonRaker restack.
• RemoteSnapmaker.cpp still captures engine by shared_ptr in SetConnectionFailureCallback — Edge-only Ultra path, not in Fix MQTT disconnect crash: lifetime & race fixes in MQTT/MoonRaker/HttpServer Snapmaker/OrcaSlicer#870's file list.

Screenshots/Recordings/Graphs

N/A — concurrency / lifetime fix. No UI change.

Tests

• Static residual diff vs Snapmaker Fix MQTT disconnect crash: lifetime & race fixes in MQTT/MoonRaker/HttpServer Snapmaker/OrcaSlicer#870: MQTT core markers already satisfied (table above); only HttpServer start_locked + SSWCP engine weak_ptr applied.
• Confirmed MQTT.cpp/.hpp have zero diff.
• Confirmed Ultra login handlers and SSWCP unload/archive paths are unchanged.
• Full slicer binary not built here (no deps/AppImage toolchain). Health-check restart path now exercises start_locked() under m_server_mtx with start_health_check() after unlock, so is_healthy() cannot observe the torn-down gap.

Still owed on real hardware (not run in this environment):

• Repeated MQTT WAN connect/disconnect: no crash
• LAN MoonRaker connect/disconnect: no crash
• App shutdown with device page open and in-flight HTTP: no crash
• Health-check restart path: no is_healthy UAF / double restart