[SYCL][PI][L0] ZeSerialize false race fix (intel#2672)

Intel Inspector race condition detector doesn't think there is a race that
can occur when writing ZeSerialize on one thread while reading it on another.
There was never really a race before because all threads would write the same
value, and thus there was no issue.  But this way of writing the code removes
the false race, and makes it obvious this global variable can only be
initialized once.

Author: kbsmith-intel

sycl/plugins/level_zero/pi_level_zero.cpp

@@ -36,7 +36,12 @@ enum {
   ZeSerializeBlock =
       2, // blocking ZE calls, where supported (usually in enqueue commands)
 };
-static pi_uint32 ZeSerialize = 0;
+static const pi_uint32 ZeSerialize = [] {
+  const char *SerializeMode = std::getenv("ZE_SERIALIZE");
+  const pi_uint32 SerializeModeValue =
+      SerializeMode ? std::atoi(SerializeMode) : 0;
+  return SerializeModeValue;
+}();
 
 // This class encapsulates actions taken along with a call to Level Zero API.
 class ZeCall {
@@ -693,11 +698,6 @@ pi_result piPlatformsGet(pi_uint32 NumEntries, pi_platform *Platforms,
       ZeValidationLayer = true;
   }
 
-  static const char *SerializeMode = std::getenv("ZE_SERIALIZE");
-  static const pi_uint32 SerializeModeValue =
-      SerializeMode ? std::atoi(SerializeMode) : 0;
-  ZeSerialize = SerializeModeValue;
-
   if (NumEntries == 0 && Platforms != nullptr) {
     return PI_INVALID_VALUE;
   }