Fix flaky uplc-evaluator integration tests (#7557)

Two issues caused intermittent test failures on CI:

1. Race condition in result file reading: Tests detected file existence
   before service finished writing, causing JSON parse errors
   ("Unexpected end-of-input"). Fixed by using atomic writes - write to
   a temp file first, then rename to final path.

2. Overly strict timing assertions: Very fast evaluations (especially
   in CI environments) can complete in <1ns, causing cpu_time_ns=0.
   Relaxed assertions from `> 0` to `>= 0` since 0ns is legitimate
   for trivial programs.

Author: Unisay

plutus-benchmark/uplc-evaluator/Main.hs

@@ -398,14 +398,16 @@ processProgram Config {..} inputPath = do
         `catch` \(e :: SomeException) ->
           hPutStrLn stderr $ "Failed to rename file: " ++ show e
 
--- | Write successful result to JSON file
+-- | Write successful result to JSON file (atomic write via temp file)
 writeResult :: FilePath -> EvalResult -> IO ()
 writeResult outputDir result = do
   let outputPath = outputDir </> T.unpack (erProgramId result) ++ ".result.json"
-  BSL.writeFile outputPath (Aeson.encode result)
+      tempPath = outputPath ++ ".tmp"
+  BSL.writeFile tempPath (Aeson.encode result)
+  renameFile tempPath outputPath
   hPutStrLn stderr $ "Wrote result: " ++ outputPath
 
--- | Write error result to JSON file
+-- | Write error result to JSON file (atomic write via temp file)
 writeError :: FilePath -> Text -> Text -> Text -> IO ()
 writeError outputDir programId errorType errorMessage = do
   let evalError =
@@ -416,7 +418,9 @@ writeError outputDir programId errorType errorMessage = do
           , eeErrorMessage = errorMessage
           }
       outputPath = outputDir </> T.unpack programId ++ ".error.json"
-  BSL.writeFile outputPath (Aeson.encode evalError)
+      tempPath = outputPath ++ ".tmp"
+  BSL.writeFile tempPath (Aeson.encode evalError)
+  renameFile tempPath outputPath
   hPutStrLn stderr $ "Wrote error: " ++ outputPath
 
 -- | Main evaluation loop

plutus-benchmark/uplc-evaluator/test/Spec.hs

@@ -98,13 +98,14 @@ main = withUtf8 do
                     ("timing_samples should have 10-20 entries, got " ++ show sampleCount)
                     (sampleCount >= 10 && sampleCount <= 20)
 
-                  -- Verify each timing sample has positive cpu_time_ns
+                  -- Verify each timing sample has non-negative cpu_time_ns
+                  -- Note: Very fast evaluations can report 0ns due to measurement resolution
                   mapM_
                     ( \s -> do
-                        -- Check that cpu_time_ns is in reasonable range
+                        -- Check that cpu_time_ns is in reasonable range (>= 0, bounded above)
                         assertBool
-                          ("cpu_time_ns should be > 0, got " ++ show (tsCpuTimeNs s))
-                          (tsCpuTimeNs s > 0)
+                          ("cpu_time_ns should be >= 0, got " ++ show (tsCpuTimeNs s))
+                          (tsCpuTimeNs s >= 0)
                     )
                     (erTimingSamples result)
 
@@ -555,13 +556,13 @@ main = withUtf8 do
                     (erMemoryBytes result > 0)
 
                   -- cpu_time_ns values may vary (timing is non-deterministic)
-                  -- We just verify they exist and are positive
+                  -- We verify they exist and are non-negative (can be 0 for very fast evaluations)
                   let cpuTimes = map tsCpuTimeNs samples
                   mapM_
                     ( \t ->
                         assertBool
-                          ("cpu_time_ns should be > 0, got " ++ show t)
-                          (t > 0)
+                          ("cpu_time_ns should be >= 0, got " ++ show t)
+                          (t >= 0)
                     )
                     cpuTimes
             ]