[AMD][CI] Swap the AMD PR gate to ROCm 7.2 and demote ROCm 7.0 to a daily shadow (#34204)

Co-authored-by: yctseng0211 <yctseng@smci355-ccs-aus-m12-17.cs-aus.dcgpu>
Co-authored-by: Chen <bingxche@amd.com>

.github/workflows/pr-test-amd.yml

@@ -1,19 +1,16 @@
-name: PR Test (AMD)
+name: PR Test ROCm 7.0 (AMD)
 # Dynamic run-name for /rerun-stage commands to enable URL lookup
 # Format: "[stage-name] sha" for fork PRs, "[stage-name]" for non-fork, default for normal runs
 run-name: ${{ (inputs.target_stage || inputs.target_stage_select) && (inputs.pr_head_sha && format('[{0}] {1}', inputs.target_stage || inputs.target_stage_select, inputs.pr_head_sha) || format('[{0}]', inputs.target_stage || inputs.target_stage_select)) || '' }}
 
 on:
   schedule:
-    - cron: '0 */12 * * *' # Run every 12 hours (UTC)
-  pull_request:
-    paths:
-      - "python/**"
-      - "scripts/ci/**"
-      - "test/**"
-      - "python/sglang/kernels/aot/**"
-      - ".github/workflows/pr-test-amd.yml"
-      - "docker/rocm.Dockerfile"
+    # ROCm 7.0 rollback shadow. pr-test-amd-rocm720.yml is the PR gate; this
+    # workflow now only proves ROCm 7.0 still works, so it fires once a day
+    # alongside nightly-test-amd.yml and chases coverage rather than latency:
+    # no matrix parallelism caps, no stage fast-fail (the `schedule` escape in
+    # each job's `if`), and continue-on-error via check-changes.
+    - cron: '30 17 * * *'
   workflow_dispatch:
     inputs:
       target_stage_select:
@@ -223,6 +220,7 @@ jobs:
     with:
       ref: ${{ inputs.pr_head_sha || inputs.ref || '' }}
       runner_arch: ${{ inputs.runner_arch || 'mi300' }}
+      rocm_version: rocm700
       aiter_ref: ${{ inputs.aiter_ref }}
       continue_on_error: true
     secrets: inherit
@@ -479,7 +477,6 @@ jobs:
       )
     strategy:
       fail-fast: false
-      max-parallel: 4
       matrix:
         part: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]
     runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -595,7 +592,6 @@ jobs:
       )
     strategy:
       fail-fast: false
-      max-parallel: 1
       matrix:
         part: [0, 1, 2]
     runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -636,7 +632,6 @@ jobs:
       )
     strategy:
       fail-fast: false
-      max-parallel: 1
       matrix:
         part: [0, 1]
     runs-on: ${{ format('linux-{0}-2gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -677,7 +672,6 @@ jobs:
       )
     strategy:
       fail-fast: false
-      max-parallel: 1
       matrix:
         part: [0, 1, 2, 3]
     runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -808,7 +802,6 @@ jobs:
       )
     strategy:
       fail-fast: false
-      max-parallel: 1
       matrix:
         part: [0, 1, 2]  # 3 partitions: 2 parametrized + 1 standalone (single_test_file/test_disagg_server.py)
     runs-on: ${{ format('linux-{0}-2gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -1082,7 +1075,6 @@ jobs:
       RUNNER_LABELS: ${{ format('linux-{0}-8gpu-sglang', inputs.runner_arch || 'mi300') }}
     strategy:
       fail-fast: false
-      max-parallel: 2
       matrix:
         part: [0, 1, 2, 3]
     runs-on: ${{ format('linux-{0}-8gpu-sglang', inputs.runner_arch || 'mi300') }}