test: exclude flaky wall-clock performance tests from CI and default test runs

Timing-based tests (CachedDiscoveryPerformanceTests, cancellation promptness
checks) are inherently flaky under CI load. Tag them with
[Trait("Category", "Benchmark")] and filter them out of the default
`dotnet test` run in CI and the /run-tests skill. They can still be run
explicitly with `--filter "Category=Benchmark"`.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: antonsynd

.claude/skills/run-tests/SKILL.md

@@ -28,8 +28,8 @@ Run Sharpy tests with optional filter. Output is smart-truncated to avoid token
 1. Run `mkdir -p .claude/tmp` to ensure log directory exists
 2. Clear the old log with `rm -f .claude/tmp/last-test-run.log`
 3. Build first: `dotnet build sharpy.sln --nologo -v q >> .claude/tmp/last-test-run.log 2>&1`. If build fails, print "=== BUILD FAILED — cannot run tests ===" then `tail -30 .claude/tmp/last-test-run.log` and stop.
-4. If `$ARGUMENTS` is non-empty, run: `dotnet test --filter "FullyQualifiedName~$ARGUMENTS" --no-build --logger "console;verbosity=normal" >> .claude/tmp/last-test-run.log 2>&1`
-5. Otherwise run: `dotnet test --no-build --logger "console;verbosity=normal" >> .claude/tmp/last-test-run.log 2>&1`
+4. If `$ARGUMENTS` is non-empty, run: `dotnet test --filter "FullyQualifiedName~$ARGUMENTS&Category!=Benchmark" --no-build --logger "console;verbosity=normal" >> .claude/tmp/last-test-run.log 2>&1`
+5. Otherwise run: `dotnet test --filter "Category!=Benchmark" --no-build --logger "console;verbosity=normal" >> .claude/tmp/last-test-run.log 2>&1`
 6. Check exit code:
    - Exit 0: Print "=== TESTS PASSED ===" then `tail -20 .claude/tmp/last-test-run.log`
    - Exit non-zero: Print "=== TESTS FAILED (last 80 lines) ===" then `tail -80 .claude/tmp/last-test-run.log`, then echo "Full log: .claude/tmp/last-test-run.log"

.github/workflows/dotnet10.yml

@@ -43,4 +43,4 @@ jobs:
     - name: Build
       run: dotnet build --no-restore
     - name: Test
-      run: dotnet test --no-build --verbosity normal
+      run: dotnet test --no-build --verbosity normal --filter "Category!=Benchmark"

src/Sharpy.Compiler.Tests/CancellationTests.cs

@@ -64,6 +64,7 @@ public void Compile_WithAlreadyCancelledToken_ThrowsOperationCanceledException()
     }
 
     [Fact]
+    [Trait("Category", "Benchmark")]
     public void Compile_WithAlreadyCancelledToken_ReturnsQuickly()
     {
         // Arrange
@@ -85,6 +86,7 @@ public void Compile_WithAlreadyCancelledToken_ReturnsQuickly()
     }
 
     [Fact]
+    [Trait("Category", "Benchmark")]
     public void Compile_WithDelayedCancellation_TerminatesPromptly()
     {
         // Arrange - generate a large program that takes nontrivial time to compile

src/Sharpy.Compiler.Tests/CancellationTokenComponentTests.cs

@@ -59,6 +59,7 @@ public void Lexer_PreCancelledToken_WithSourceText_ThrowsDuringTokenizeAll()
     }
 
     [Fact]
+    [Trait("Category", "Benchmark")]
     public void Lexer_PreCancelledToken_LargeSource_TerminatesPromptly()
     {
         // Even with a large source, a pre-cancelled token should terminate quickly
@@ -137,6 +138,7 @@ public void Parser_PreCancelledToken_SmallProgram_CompletesWithoutCrash()
     }
 
     [Fact]
+    [Trait("Category", "Benchmark")]
     public void Parser_PreCancelledToken_LargeTokenList_TerminatesPromptly()
     {
         // Generate a large program and tokenize it
@@ -214,6 +216,7 @@ public void Compile_PreCancelledToken_DoesNotProduceGeneratedCode()
     }
 
     [Fact]
+    [Trait("Category", "Benchmark")]
     public void Compile_CancelledDuringLargeProgram_TerminatesPromptly()
     {
         // Use a token that cancels after a short delay

src/Sharpy.Compiler.Tests/Performance/CachedDiscoveryPerformanceTests.cs

@@ -12,6 +12,7 @@ namespace Sharpy.Compiler.Tests.Performance;
 /// <summary>
 /// Performance benchmarks for the cached overload discovery system
 /// </summary>
+[Trait("Category", "Benchmark")]
 public class CachedDiscoveryPerformanceTests : IDisposable
 {
     private readonly ITestOutputHelper _output;