CBG-5460: Re-enable windows toplogy tests

[gregns1]

CBG-5460

Adds change to windows builds to get HLC time with GetSystemTimePreciseAsFileTime, reading the high-resolution system clock live on every call.

The Hybrid Logical Clock derives a version's physical component from the wall clock, then clears the low 16 logical bits — making each physical "slot" ~65 µs wide. On Linux/macOS, time.Now() resolves to ~1 ns, so this is a non-issue. On Windows, time.Now() is backed by the coarse system timer (~0.5–15 ms). In topology tests, peers writing in quick succession produced identical HLC versions, because many writes fell inside a single coarse tick.

The initial fix used QueryPerformanceCounter (QPC): snapshot a (QPC ticks, time.Now()) anchor pair once at init(), then compute anchorNanos + elapsedTicks. This resolves to ~100 ns which resolves the resolution issue.

The two clocks ended up with a fixed, never-correcting offset of up to ~15 ms between them — and because it depends on startup timing, the offset is random per process run. Constantly tripping the cv.ver <= cas re-stamp path and producing intermittent topology-test failures.

Switched to use GetSystemTimePreciseAsFileTime which fixes the issues as offers same precision but both base and rosmar read the same system clock

Sync Gateway's HLC and Couchbase Server's CAS are genuinely independent clocks on separate nodes, and the system is designed to converge under that skew — the CAS re-stamp exists precisely for it. When the topology test failed, the data still converged correctly (every peer agreed on the same cv and pv); only the test harness's predicted version was wrong.

Pre-review checklist

• Removed debug logging (fmt.Print, log.Print, ...)
• Logging sensitive data? Make sure it's tagged (e.g. base.UD(docID), base.MD(dbName))
• Updated relevant information in the API specifications (such as endpoint descriptions, schemas, ...) in docs/api

Dependencies (if applicable)

• CBG-5460 add windows precision for HLC couchbaselabs/rosmar#79
• Update Go module dependencies when merged

Integration Tests

• https://jenkins.sgwdev.com/job/SyncGatewayIntegration/827/
• https://jenkins.sgwdev.com/job/TopologyTests/378

[Copilot]

Pull request overview

This PR aims to make HLV-based topology tests deterministic on Windows by spacing peer writes to avoid ties caused by Windows’ lower wall-clock precision (independent HLCs producing identical versions).

Changes:

• Add Windows-only time.Sleep delays before peer document mutations (create/update/delete) to help ensure unique HLV versions across peers.
• Introduce runtime/time usage in the topology HLV test helpers to gate this behavior by OS.

[torcolvin]

As discussed offline, see if we can use QueryPerformanceCounter to get a high resolution timestamp to replace for use in HybridLogicalClock. https://learn.microsoft.com/en-us/windows/win32/sysinfo/acquiring-high-resolution-time-stamps

[torcolvin]

I like this fix but I think I'd like to move the hlc implementation to sg-bucket since we use this in rosmar too!

[torcolvin]

@bbrks can you have a look at this review? This is just uptaking the change from sg-bucket. I originally reviewed this code in sync_gateway that Greg did and then I merged this to sg-bucket.