Set CHPL_TARGET_CPU=native in GPU performance testing (chapel-lang#24281

)

We got confused by lack of this setting in the past. Without it, runs
always generate warnings that get ignored in the performance setting.
That is fine in steady state, but if there's a failure, we see the
warning in the full output, which is almost always a red herring.

[Trivial, not reviewed]

util/cron/test-perf.gpu-cuda.bash

@@ -20,6 +20,7 @@ source $CWD/common-native-gpu-perf.bash
 # make sure this comes after setting SUBDIR (set by native-gpu-perf) and
 # CONFIG_NAME
 source $CWD/common-perf.bash
+export CHPL_TARGET_CPU=native
 
 
 nightly_args="${nightly_args} -startdate 07/15/22"

util/cron/test-perf.gpu-cuda.um.bash

@@ -21,6 +21,7 @@ source $CWD/common-native-gpu-perf.bash
 # make sure this comes after setting SUBDIR (set by native-gpu-perf) and
 # CONFIG_NAME
 source $CWD/common-perf.bash
+export CHPL_TARGET_CPU=native
 
 SHORT_NAME=um
 nightly_args="${nightly_args} -performance-description $SHORT_NAME -performance-configs default:v,$SHORT_NAME:v -sync-dir-suffix $SHORT_NAME"

util/cron/test-perf.gpu-rocm.bash

@@ -18,6 +18,7 @@ source $CWD/common-native-gpu-perf.bash
 # make sure this comes after setting SUBDIR (set by native-gpu-perf) and
 # CONFIG_NAME
 source $CWD/common-perf.bash
+export CHPL_TARGET_CPU=native
 
 # everything we source above will end up sourcing `common.bash` which will then
 # source `load-base-deps.bash`. In the system we run this config,