feat(net): add retry with backoff to network calls

[aiexkwan]

Summary

• PetBrowser.fetch: add 10s timeout + 1 retry for transient failures (429/5xx/timeout) — previously had no timeout and silently returned []
• CareSyncController.push: add 15s timeout + exponential backoff (30→120→300→600s) + cap at 5 consecutive failures with user-facing message — previously used fixed 300s retry with no limit
• NativeUsageProbe: retry once on transient 5xx for both Claude and Codex probes — previously a single server hiccup would drop the provider until next poll cycle

Follows the existing PetdexAssets inline retry pattern (no new utility classes). OpenUsageClient (local 127.0.0.1 endpoint) and PetInstaller (already has retry) are intentionally unchanged.

Test plan

• swift build — compiles cleanly
• swift test — 194 tests, 0 failures, no regressions
• Manual: disconnect Wi-Fi briefly while opening Browse Pets → pet library should recover after reconnecting (retry fires after 1s)
• Manual: observe Care tab sync status after simulated server error → should show "Sync failed, will retry." with increasing delay

🤖 Generated with Claude Code

[ntd4996]

Merged and shipped in v1.5.0. Solid hardening, the timeouts + backoff and the consecutive-failure cap on Care sync are a real improvement over the old fixed 300s retry, and reusing the existing inline retry pattern keeps it clean. Thanks!