fix: Resolve flaky TestStore_Integration spurious failures

[corylanou]

Summary

Fixes a race condition in TestStore_Integration that was causing spurious test failures at exactly 10.01s.

Root Cause

The test has a goroutine race condition where the insert goroutine calls t.Errorf() after the main test has completed:

1. Main test goroutine: Exits immediately when 10s timer closes the done channel
2. Insert goroutine: May still be mid-operation using t.Context()
3. Race condition: When main test exits, t.Context() gets cancelled
4. Spurious failure: Insert goroutine gets "database locked" error and calls t.Errorf() after test completion

In Go 1.14+, calling testing methods from a goroutine after the test has finished causes the test to fail.

Evidence

• Failed CI run: https://github.com/benbjohnson/litestream/actions/runs/17583602203/job/49945685558
• Failure pattern: Always fails at exactly 10.01s (just after the 10s test duration)
• Error logs: store_test.go:150: insert error: database is locked (5) (SQLITE_BUSY)

Solution

• Error channel: Capture insert errors from the goroutine
• Shutdown awareness: Check if done channel is closed before treating errors as failures
• Defer cleanup: Use defer to check error channel on any test exit path
• Expected behavior: Errors during shutdown (after done closes) are ignored as expected

Test Results

The fix eliminates spurious failures while still catching legitimate insert errors that occur during normal test execution.

# Before: Intermittent failures at 10.01s
--- FAIL: TestStore_Integration (10.01s)

# After: Reliable passes  
--- PASS: TestStore_Integration (10.18s)

Changes

• Add insertErr channel to capture errors from insert goroutine
• Add shutdown-aware error handling: only report errors if not shutting down
• Use defer to ensure error checking happens on any test exit
• Maintain original clean loop structure with return statements

🤖 Generated with Claude Code