Fix the source-lane shell dispatch; pin cudaq to the 0.16.0 release

.cudaq_version

@@ -1,6 +1,6 @@
 {
   "cudaq": {
     "repository": "NVIDIA/cuda-quantum",
-    "ref": "b28cf0f6f2d9387f12ca81b6824b8833a38530af"
+    "ref": "51671baac373c545c9651c3cfa1a8d371aa67ec7"
   }
 }

.github/workflows/lib_algorithms.yaml

@@ -52,9 +52,21 @@ on:
         required: false
 
 concurrency:
-  group: ${{ github.workflow }}-build-algorithms-${{ github.ref }}
+  # cudaq_source must be part of the group: one PR run can invoke this
+  # workflow twice (pip lanes + pin-triggered source lanes), and a
+  # shared group would have the later invocation cancel the earlier.
+  group: ${{ github.workflow }}-build-algorithms-${{ inputs.cudaq_source }}-${{ github.ref }}
   cancel-in-progress: true
 
+# The build-and-test job runs inside the cuda-quantum devcontainer, where
+# the runner's default shell for `run:` steps resolves to `sh` (dash).
+# The steps below use bash syntax (`[[ ]]`), which dash rejects — and a
+# failing `[[` inside an `if` condition silently selects the else branch
+# instead of erroring the step. Pin bash explicitly.
+defaults:
+  run:
+    shell: bash
+
 jobs:
   build-and-test:
     name: Build and test

.github/workflows/wheel_algorithms.yaml

@@ -58,7 +58,10 @@ on:
         required: false
 
 concurrency:
-  group: ${{ github.workflow }}-build-wheels-${{ github.ref }}
+  # cudaq_wheels must be part of the group: a pin change invokes this
+  # workflow in both PyPI and Custom modes within one PR run, and a
+  # shared group would cancel the first with the second.
+  group: ${{ github.workflow }}-build-wheels-${{ inputs.cudaq_wheels }}-${{ github.ref }}
   cancel-in-progress: true
 
 jobs: