Expose upload-health metrics (success/failure/retry/last-success) on the FinOps Agent's Prometheus /metrics endpoint

Container costs are a core part of our overall cloud spend, so we depend on the FinOps Agent uploading reliably to Cloudability. Today the agent's /metrics endpoint (metrics.enabled=true) only exposes generic Go runtime/process metrics (go_, process_) - there is no counter, gauge, or timestamp anywhere that reflects whether Cloudability uploads are actually succeeding.

This creates a real blind spot: the pod can be Running/Ready with 0 restarts (so kube-state-metrics and any standard Kubernetes health check report it as fully healthy) while Frontdoor authentication or the S3 upload step fails continuously in the background. The only way to detect this today is to manually read pod logs or wait for data to appear (or fail to appear) in the Cloudability UI, which can take 24-48 hours to reflect an outage. That delay is too long for an operationally useful alert, and manual log inspection isn't a substitute for automated monitoring at scale.

We looked at working around this ourselves (e.g. adding a sidecar to tail the agent's logs and turn success/failure lines into metrics) but decided against it. We'd rather rely on the agent's own out-of-the-box observability story than build and maintain custom, chart-specific plumbing that fights against the supported deployment model. We suspect most operators running this agent in production, especially anyone treating container cost visibility as business-critical, would run into the exact same gap and reach the same conclusion.

Concretely, we'd like the existing Cloudability emitter to expose (behind the current metrics.enabled flag, consistent with how the collector data source is already instrumented):

• A counter for successful uploads (e.g. cloudability_upload_success_total)

• A counter for failed uploads (e.g. cloudability_upload_failure_total), ideally with a reason/error-type label

• A counter or gauge for retry attempts (e.g. cloudability_upload_retry_total), since agent.cloudability.uploadRetryCount already exists as a config knob but its outcomes aren't observable

• A gauge for the last successful upload timestamp (e.g. cloudability_last_successful_upload_timestamp_seconds), so a "time since last success" alert is possible

This would let us (and any other Prometheus/VictoriaMetrics-based shop) write a simple, standard alerting rule that fires within minutes of a real upload failure, instead of discovering it a day or two later through the UI. This is a small, additive change to an emitter path that's already configured and running - it doesn't require new agent capabilities, just instrumenting the upload path the same way the collector path is already instrumented.