geoprobe-agent: filter cached DZD measurements to avoid jittery measurements

[ben-dz]

Resolves: #3219

Summary of Changes

• Replace single-entry-per-sender offset cache with a two-slot (best + backup) strategy to prevent latency spikes when the best measurement expires
• best slot holds the lowest RTT offset across the full validity window; backup holds the lowest RTT from the recent half-window, ensuring a smooth handoff when best expires
• Higher-RTT offsets never overwrite a lower-RTT best; backup is refreshed when stale (older than half-maxAge) to guarantee remaining lifespan

Diff Breakdown

Category	Files	Lines (+/-)	Net
Core logic	1	+71 / -18	+53
Tests	1	+162 / -1	+161

Compact change — ~50 net lines of core logic, well-covered by 7 new test cases.

Key files (click to expand)

• controlplane/telemetry/cmd/geoprobe-agent/main.go — new cachedSender struct with best/backup slots; rewritten Put() with RTT filtering and backup promotion, Get() with fallback to backup, GetBest() iterating both slots, Evict() checking both slots
• controlplane/telemetry/cmd/geoprobe-agent/main_test.go — 7 new tests (PutKeepsBest, PutLowerRTTReplacesBest, BackupPromotion, BackupRefreshInHalfWindow, GetBestWithTwoSlots, EvictBothSlots) + updated PutReplaces expectation

Testing Verification

• All 27 unit tests pass (go test -v -count=1 ./controlplane/telemetry/cmd/geoprobe-agent/...)
• Backup promotion verified: best expires → backup takes over without spike
• Half-window refresh verified: backup refreshed past half-maxAge so it has remaining lifespan when best expires
• RTT filtering verified: worse measurements stored as backup, never overwrite best
• Concurrent access test exercises all methods under contention