Fix diags.chpl gpu test (add separate good file for cpu-as-device) (c…

…hapel-lang#23698)

In a prior PR (chapel-lang#22592) I,
among other things, changed some allocations during kernel launch to use
`chpl_mem_alloc` instead of `chpl_malloc`. The effect of this is these
allocations are now being reported in our memory diagonstics log when
previously they weren't. I updated the `.good` file associated with the
`diags.chpl` to account for this but didn't account for the fact that
these newly reported allocations won't happen when using cpu-as-device.

In this PR I add a new `.cpu.good` file for when we do cpu-as-device
testing and update the `.execopts` file to be a script that will tell
start_test to use this new good file when appropriate.

[Reviewed by: nobody; update to nightly test]

test/gpu/native/diags.cpu.good

@@ -0,0 +1,13 @@
+warning: The prototype GPU support implies --no-checks. This may impact debuggability. To suppress this warning, compile with --no-checks explicitly
+Start
+0 (cpu): diags.chpl:nn: allocate xxB of domain(1,int(64),one) at 0xPREDIFFED
+0 (cpu): diags.chpl:nn: allocate xxB of [domain(1,int(64),one)] int(64) at 0xPREDIFFED
+0 (cpu): diags.chpl:nn: allocate xxB of array elements at 0xPREDIFFED
+0 (cpu): diags.chpl:nn: allocate xxB of _EndCount(atomic int(64),int(64)) at 0xPREDIFFED
+0 (cpu): diags.chpl:nn: free xxB of _EndCount(atomic int(64),int(64)) at 0xPREDIFFED
+0 (gpu 0): diags.chpl:nn: allocate xxB of [domain(1,int(64),one)] int(64) at 0xPREDIFFED
+0 (gpu 0): diags.chpl:nn: allocate xxB of array elements at 0xPREDIFFED
+0 (gpu 0): diags.chpl:nn: free xxB of array elements at 0xPREDIFFED
+0 (gpu 0): diags.chpl:nn: free xxB of [domain(1,int(64),one)] int(64) at 0xPREDIFFED
+2 2 2 2 2 2 2 2 2 2
+End

test/gpu/native/diags.execopts

@@ -1 +1,7 @@
---memTrack
+#!/usr/bin/env bash
+
+if [[ "$CHPL_GPU" == "cpu" ]]; then
+  echo "--memTrack # diags.cpu.good"
+else
+  echo "--memTrack"
+fi