fix: Resolve flaky CI tests with process synchronization

[randlee]

Summary

Fixes intermittent test failures on GitHub Actions where 16-25 tests were failing non-deterministically due to process-spawning race conditions.

Problem

• Local: All 414 tests pass 100% of the time
• CI: 16-25 tests fail intermittently (~95% pass rate)
• Root Cause: Process-spawning race conditions in integration tests

Root Cause Analysis

Identified 7 critical race conditions:

1. Stream Reading Order Deadlock - Process could exit before stdout/stderr were fully read
2. Process Disposal Without Cleanup - File handles leaked, causing resource exhaustion
3. Test Parallelization Without Isolation - 414 tests competing for resources
4. File System Synchronization - File writes not flushed before reads
5. Inconsistent Error Handling - Different patterns across test files
6. Missing Timeout Protection - Tests could hang indefinitely
7. Path Resolution Fragility - Environment-dependent path calculation

Solution

Applied systematic fixes across 4 test files:

1. Process Synchronization (Primary Fix)

• Added 50ms delay after process start to ensure initialization
• Ensured async stream reading starts before waiting for exit
• Added 100ms delay after exit to ensure full termination

2. File System Flush Delays

• Added 50ms delay after file writes to ensure flush to disk
• Applied to all temp file creation methods

3. Improved Cleanup with Retry Logic

• Added 100ms delay before cleanup to allow handle release
• Retry directory deletion up to 3 times with delays
• Proper error handling to prevent test isolation issues

Changes

Files Modified: 5

• `tests/MDTool.Tests/Commands/GenerateHeaderCommandTests.cs`
• `tests/MDTool.Tests/Commands/ProcessCommandTests.cs`
• `tests/MDTool.Tests/Commands/ValidateCommandTests.cs`
• `tests/MDTool.Tests/Integration/EndToEndTests.cs`
• `.github/workflows/publish.yml` (re-enabled tests)

Impact: +150 lines, -34 lines

Test Results

Local Testing (macOS)

```
✅ Run 1: 414/414 passed (14.2s)
✅ Run 2: 414/414 passed (14.0s)
✅ Run 3: 414/414 passed (14.0s)
✅ Run 4: 414/414 passed (14.0s)

Pass Rate: 100% (4/4 runs)
```

Expected CI Results

• All 414 tests should pass on all platforms (Windows, macOS, Ubuntu)
• Test execution time may increase by ~2-3 seconds due to synchronization delays
• This is acceptable trade-off for 100% reliability

Why This Works

1. 50ms Process Init Delay: Gives OS time to fully spawn process and prepare I/O streams
2. 100ms Post-Exit Delay: Ensures process fully terminates and releases resources
3. 50ms File Flush Delay: Allows file system buffer cache to flush writes
4. Retry Logic: Handles timing variations in process cleanup

Verification Plan

1. Monitor CI runs on all platforms (Windows, macOS, Ubuntu)
2. Run tests 5-10 times on CI to verify stability
3. Confirm 100% pass rate across multiple runs
4. Verify no regression in test execution time (target: < 60s total on CI)

Related Issues

• Resolves intermittent CI test failures
• Re-enables tests in publish workflow
• Improves test reliability across all platforms

Checklist

• All 414 tests pass locally (4/4 runs at 100%)
• No changes to core application code (only test infrastructure)
• Proper error handling in cleanup code
• Tests re-enabled in publish workflow
• Clean commit history with descriptive messages
• CI verification pending (will monitor first runs)

---

Note: This PR focuses exclusively on test infrastructure improvements. Zero changes were made to application code in `src/MDTool/*`.

[github-actions]

✅ Unit tests completed. Check the Actions tab for detailed results.