vm/cc/Profiler_BinaryOperatorSourcePosition flakily fails with `profiler.cc: 111: error: expected: sample_buffer_ != nullptr`

[mkustermann]

From this log:

/=====================================================================================================================\
| dartk-weak-asserts-linux-debug-x64:vm/cc/Profiler_BinaryOperatorSourcePosition broke (Pass -> Crash, expected Pass) |
\=====================================================================================================================/

--- Command "run_vm_unittest" (took 02.000806s):
DART_CONFIGURATION=DebugX64 out/DebugX64/run_vm_tests --dfe=/b/s/w/ir/out/DebugX64/gen/kernel-service.dart.snapshot --enable-experiment=records --ignore-unrecognized-flags Profiler_BinaryOperatorSourcePosition

exit code:
-6

stdout:
Running test: Profiler_BinaryOperatorSourcePosition
Done: Profiler_BinaryOperatorSourcePosition

stderr:
../../runtime/vm/profiler.cc: 111: error: expected: sample_buffer_ != nullptr

===== CRASH =====
si_signo=Segmentation fault(11), si_code=1, si_addr=0xababb22b
Aborting reentrant request for stack trace.

--- Re-run this test:
python3 tools/test.py -n dartk-weak-asserts-linux-debug-x64 vm/cc/Profiler_BinaryOperatorSourcePosition

[bkonyi]

I haven't been able to reproduce this running under load for an hour. It looks like this is failing during cleanup though, so maybe we just need an extra null check somewhere.