perf(tunnel): reduce telemetry bandwidth for v0.17.6

[youzi-1122]

Summary

• Negotiate Snappy compression for metric RPC payloads while preserving JSON compatibility with older Managers and Edge agents.
• Split large metric uploads within Geminio's real frame budget, including Base64 envelope overhead, and retain acknowledged progress when later batches fail.
• Avoid re-uploading confirmed cached scrape snapshots and duplicate scrape URLs without dropping new observations or changing their timestamps.
• Bump VERSION and all README download examples to v0.17.6. Helm chart metadata and Edge image tags are derived from this version during packaging.

Design and compatibility details: docs/rfc/RFC-004-tunnel-metrics-compression.md.

Validation

• Linux arm64 race tests passed for tunnel, collectors, Edge business logic, metrics, custom metrics, database metrics, Auto APM, Kubernetes, and Manager Frontier handlers.
• Changed-package go vet and Linux builds of cmd/ongrid and cmd/ongrid-edge passed.
• Regression tests cover JSON fallback, bounded decoding, partial batch acknowledgement, cancellation, snapshot deduplication, and actual Geminio packet encoding below its 10 MiB frame limit.
• A synthetic multi-batch payload retained all samples and reduced 9,436,686 JSON bytes to 442,915 Snappy bytes. This is not a production bandwidth estimate.
• make proto, make version-print package-k8s-chart, and git diff --check passed; packaged chart version, appVersion, and image tag were verified as 0.17.6 / v0.17.6 / v0.17.6.

Rollout and rollback

Upgrade Manager before Edge to enable compression immediately; mixed versions retain the JSON path when compression is not negotiated. Roll back to v0.17.5 artifacts if necessary; there are no database migrations. Durable buffering and exactly-once delivery are outside this change, and production load testing has not been performed.

Author confirmation

• I have read and agree to the project contribution guide. Guide: https://github.com/ongridio/ongrid/blob/main/CONTRIBUTING.md