[Windows] Reduce trunk model test count (pytorch#14348)

Reduce the number of models we test on Windows CI in order to reduce
runner utilization. I've tried to pick a reasonably representative set.
I don't think we're getting much incremental value from the ones we're
cutting and will save on CI costs. We can re-evaluate this when we have
shared/cached build for Windows CI runs.

Author: GregoryComer

.github/workflows/trunk.yml

@@ -1016,8 +1016,8 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        model: [linear, add, add_mul, ic3, ic4, mv2, mv3, resnet18, resnet50, vit, w2l, mobilebert, emformer_join, emformer_transcribe]
-        backend: [portable, xnnpack-f32, xnnpack-q8]
+        model: [mv3, resnet50, vit, mobilebert, emformer_transcribe]
+        backend: [portable, xnnpack-q8]
     with:
       submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}