[BE] Remove confusing opbenchmark-on-demand-build (#165583)

As it doesn't have a test shard, so what's the point or running the build? Was added in https://github.com/pytorch/pytorch/pull/143733 and looks like test shard never existed for it

Moreover, allow one to specify benchmark size as argument, so one
technically can do a workflow dispatch with different opbenchmark sizes
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165583
Approved by: https://github.com/huydhn

.github/workflows/operator_benchmark.yml

@@ -7,9 +7,11 @@ on:
   workflow_dispatch:
     inputs:
       test_mode:
-        required: false
-        type: string
-        default: 'short'
+        type: choice
+        options:
+          - 'short'
+          - 'long'
+          - 'all'
         description: tag filter for operator benchmarks, options from long, short, all
   schedule:
     # Run at 07:00 UTC every Sunday
@@ -37,20 +39,7 @@ jobs:
       docker-image-name: ci-image:pytorch-linux-jammy-py3-gcc11-inductor-benchmarks
       test-matrix: |
         { include: [
-          { config: "cpu_operator_benchmark_short", shard: 1, num_shards: 1, runner: "linux.12xlarge" },
-        ]}
-    secrets: inherit
-
-  opbenchmark-on-demand-build:
-    if: ${{ github.event_name == 'workflow_dispatch' && github.repository_owner == 'pytorch' }}
-    name: opbenchmark-on-demand-build
-    uses: ./.github/workflows/_linux-build.yml
-    with:
-      build-environment: linux-jammy-py3.10-gcc11-build
-      docker-image-name: ci-image:pytorch-linux-jammy-py3-gcc11-inductor-benchmarks
-      test-matrix: |
-        { include: [
-          { config: "cpu_operator_benchmark_${{ inputs.test_mode }}", shard: 1, num_shards: 1, runner: "linux.12xlarge" },
+          { config: "cpu_operator_benchmark_${{ inputs.test_mode || 'short' }}", shard: 1, num_shards: 1, runner: "linux.12xlarge" },
         ]}
     secrets: inherit