Fix flaky tests on Windows

[bauerpawel]

Description

Several tests in the suite fail intermittently (and in some cases deterministically) when run on Windows. Root causes:

• File handles held open across os.Remove: generateHashAndEncrypt (internal/storage/FileServing.go) and getDaysRemaining (internal/webserver/ssl/Ssl.go) didn't close a file handle before removing/rewriting that same file. On Linux this is a no-op; on Windows an open handle blocks the remove/rewrite, so the test either errors out or silently keeps stale data. Aws_test.go had the same pattern.
• chmod-based permission tests: TestGetFileByChunkId (internal/storage/chunking/Chunking_test.go) used os.Chmod to make a file read-only and assert a failing removal - POSIX permission bits are a no-op on Windows, so the assertion never held. Guarded with runtime.GOOS != "windows".
• Hardcoded /-separated path assertions: Localstorage_test.go asserted exact path strings using /, which don't match filepath-joined paths on Windows (\). Switched to os.PathSeparator.
• Wildcard vs. specific address bind conflict: TestIsErrorAddressAlreadyInUse (Setup_test.go) bound a test listener to 127.0.0.1:PORT but the second server to :PORT (wildcard). On Windows, binding to the wildcard address doesn't reliably conflict with an existing listener bound to a specific address on the same port, so the intended "address already in use" error never surfaced. Now both bind to the same specific address.
• testing/synctest vs. real I/O contention: TestParallelDownloads used testing/synctest's deterministic fake clock around 50 goroutines doing real SQLite writes. The fake clock has no way to gracefully coexist with unpredictable real-world I/O delays (e.g. antivirus briefly locking the database file) - a single external stall can look like a livelock, since retries governed by the fake clock spin far faster than the real delay actually resolves. Replaced with a plain sync.WaitGroup.

None of these are behavior changes to production code paths except the two file-handle fixes, which are strictly more correct (closing a handle before removing/rewriting the same file) on every platform.

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Technical Details

• Database changes: No
• Storage backend affected: No
• Usage of AI: Yes — developed with Claude Code as a pair-programming assistant, including root-causing the synctest livelock (confirmed via goroutine dumps that the original, unmodified code also hung, ruling out a regression). I reviewed and tested all changes before submitting.

How Has This Been Tested?

• Unit Tests: go test ./... --tags=test,awsmock (full suite passes, 15+ repeated runs of the previously-hanging TestParallelDownloads)
• Environment: Windows 11

Checklist

• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas. (added a comment explaining the synctest → WaitGroup rationale, since it's non-obvious)
• I have made corresponding changes to the documentation. (n/a - no behavior/doc change)
• My changes generate no new warnings.

🤖 Generated with Claude Code