qt: move QtWebSocketBackend onto the structural registration surface and delete the WASM special case

[Yaraslaut]

Re-scoped out of #522, step 2 of 5. Blocked by #567 — the new surface must exist first.

Why this one is the heart of the set

QtWebSocketBackend is the only backend in the tree that overrides any of the four *Async verbs (src/qt/qt_websocket_backend.cpp:160, :226, :265, :368). Every other implementor — LocalBackend, SimulatedRemoteBackend, SocketBackend, and all 11 test doubles — inherits the defaulted return false and falls back to the synchronous path.

So the entire prose threading contract in backend.hpp:131-135 exists for exactly one class, and the WASM special case it describes lives here. #522's claim that "the WASM special case disappears rather than being a documented exception" is a claim about this file.

What to do

Move QtWebSocketBackend onto the surface from #567 natively — not through SynchronousBackendAdapter, which would defeat the point. It is the one backend that genuinely has a non-blocking path, and it should be the one that demonstrates the new shape carries it.

Then delete the WASM special case rather than re-documenting it.

Acceptance

• QtWebSocketBackend implements the core: give IBackend a registration surface whose threading contract is structural, not prose #567 surface directly and overrides none of the four *Async verbs.
• The single-threaded WASM registration path works with no special case in the interface — demonstrated by the existing coverage in tests/qt/test_qt_websocket.cpp and examples/common/testkit/test_wasm_registration_path_native.cpp.
• The continuation is delivered on the thread the contract requires, and there is a test that fails if it is not. This is the morph#486 use-after-free the prose contract was guarding against; a migration that preserves the guarantee without testing it has moved the risk, not removed it.
• include/morph/qt/qt_websocket_backend.hpp's documentation describes the new shape; the prose threading contract it used to restate is gone, not copied.

Notes

First ticket in the new Q qt-gui lane (include/morph/qt, src/qt, tests/qt), created for this set. Blocks #522's step 5. The example WASM mains are not in scope here — they are step 4, in the N ladder lane.

[Yaraslaut]

Triaged as part of the #567–#571 set. Verified against origin/master.

Premise check. git grep -n 'registerModelAsync\|registerModelSharedAsync\|attachModelAsync\|assignPrimaryAsync' -- include src tests examples confirms the claim exactly: the only production definitions of these four are bool QtWebSocketBackend::registerModelAsync (src/qt/qt_websocket_backend.cpp:160), registerModelSharedAsync (:226), attachModelAsync (:265), assignPrimaryAsync (:368). LocalBackend, SimulatedRemoteBackend, and SocketBackend all inherit the defaulted return false. Read the four override bodies directly: they're gated by QtWebSocketBackendConfig::asyncRegistrationEnabled (default false), queue registrations made before _connected is true, and route replies through onTextMessage on the Qt event-loop thread — matching this ticket's and docs/spec/core/backend.md's description of the WASM special case precisely.

This ticket's premise depends on #567 landing a sound surface first — see my comment there: the proposed surface is structural (uses Completion+IExecutor, the mechanism execute() already relies on), not a relabeling, so this migration inherits a solid foundation rather than a flawed one.

No invariant is violated — this is a same-shape, same-lane consumer migration onto a surface already reviewed for soundness. Acceptance criteria (no override of the four verbs, a test that fails if the delivery thread is wrong, doc rewritten rather than re-restating the retired contract) are concrete and checkable against existing coverage (tests/qt/test_qt_websocket.cpp, examples/common/testkit/test_wasm_registration_path_native.cpp, both confirmed to exist on origin/master).

Verdict: triage: valid.

[Yaraslaut]

#585 implements this, but it cannot land as scoped: it introduces a deterministic regression in morph::net::SocketBackend (six tests/net/ cases, reproduced 5/5 locally, plus clang-ubsan/clang-tsan/Valgrind in CI).

Root cause is filed as #593. Briefly: once #568 deletes QtWebSocketBackend's four *Async overrides, Bridge's single registration call site must fall back to bindModel — and after #586 made SocketBackend::bindModel natively non-blocking, that one call site has to block for SocketBackend (whose callers use a BridgeHandler on the next line) and not block for WASM QtWebSocketBackend (where blocking is the nested-QEventLoop deadlock this ticket exists to remove).

Both horns were measured: forcing the default blocking bindModel clears all six SocketBackend failures and breaks five QtWebSocketBackend ones instead, including bindModel (non-blocking via Config::asyncRegistrationEnabled) registers without blocking. Neither setting is correct.

include/morph/core/backend.hpp:487-493 rules out the discriminator by design ("There is no bool saying 'I have no async path'"), so this needs a decision at the #567/#571 level — restore a caller-visible signal, or migrate executeVia to queue-until-bound. #568 is blocked on #593.