emdawnwebgpu: Add runtimeKeepalive for device.lost handler

The device.lost handler in emwgpuAdapterRequestDevice is wrapped in
callUserCallback(), which calls maybeExit() after the callback executes.
When runtimeKeepaliveCounter is 0 (common for --no-entry WASM modules),
maybeExit() triggers _exit(0), setting ABORT=true. This causes all
subsequent callUserCallback() invocations to silently drop their
callbacks, breaking any WebGPU operations that follow (e.g.,
requestAdapter for a new session).

Fix by adding runtimeKeepalivePush() before the device.lost promise
setup, and runtimeKeepalivePop() after callUserCallback() returns
inside the .then() handler. This ensures the runtime stays alive
during the callback and maybeExit() sees counter >= 1.

The pop is placed after callUserCallback (not before) because
callUserCallback internally calls maybeExit() - if the pop happened
first, the counter would be 0 during maybeExit, still triggering the
premature exit.

Bug: onnxruntime#27427

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: fs-eire

third_party/emdawnwebgpu/pkg/webgpu/src/library_webgpu.js

@@ -909,7 +909,9 @@ var LibraryWebGPU = {
 #if ASSERTIONS
         assert(deviceLostFutureId);
 #endif
-        // Don't keepalive here, because this isn't guaranteed to ever happen.
+        // Keep the runtime alive until device.lost resolves, to prevent
+        // maybeExit() from triggering premature ABORT during callUserCallback.
+        {{{ runtimeKeepalivePush() }}}
         WebGPU.Internals.futureInsert(deviceLostFutureId, device.lost.then((info) => {
           // If the runtime has exited, avoid calling callUserCallback as it
           // will print an error (e.g. if the device got freed during shutdown).
@@ -925,6 +927,7 @@ var LibraryWebGPU = {
               {{{ gpu.passAsPointer('messagePtr') }}});
             stackRestore(sp);
           });
+          {{{ runtimeKeepalivePop() }}}
         }));
 
         // Set up uncaptured error handlers.