add protection for potentially invalid disk data

[chadpatel]

Description of the issue

The systemmetricsreceiver disk scraper aggregates aggregate_disk_free and aggregate_disk_used by summing across all /dev/-prefixed mounts. On Linux, gopsutil's disk.UsageWithContext computes Total and Free as uint64(stat.Blocks) * uint64(stat.Bsize) from the kernel's statfs syscall. Certain filesystem states — transient loop mounts, mounts under heavy I/O pressure, or broken FUSE devices — can cause statfs to return garbage values where Free > Total or both values are near 2^63. When these are summed into the aggregate, the resulting metric is nonsensical and poisons downstream min/max rollups.

Observed in production on an EC2 host with a nearly-full root XFS volume: 5 out of ~2100 one-minute samples reported aggregate_disk_free ≈ -9.2e18 (Long.MIN_VALUE range after float64 conversion), which can break downstream consumers that don't expect negative or astronomically large byte values.

Description of changes

Adds a per-mount plausibility check (isPlausibleDiskUsage) before accumulating into the aggregate sum. A mount is rejected if:

• Total == 0 (statvfs failure or pseudo-filesystem)
• Free > Total (physically impossible — the exact signature of the observed bug)
• Total or Free exceeds 1 PiB (1 << 50 bytes, ~16x the largest EBS volume)

When any mount fails the check, the entire sample is dropped rather than emitting a partial sum. CloudWatch handles missing datapoints correctly; it does not handle wrong datapoints correctly — a partial sum silently understates free space and poisons min() aggregations.

License

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

Tests

• Added TestDiskScraperDropsSampleWhenFreeExceedsTotal — reproduces the exact bug signature (tiny total, huge free)
• Added TestDiskScraperDropsSampleWhenTotalExceedsCap — mount above 1 PiB cap
• Added TestDiskScraperDropsSampleWhenTotalIsZero — defense-in-depth (already filtered by DiskUsage())
• Added TestIsPlausibleDiskUsage — table-driven test covering 6 cases (normal, full disk, at cap, zero total, free > total, above cap)
• All existing disk scraper tests pass with Total field added to test fixtures
• Full package: go test ./receiver/systemmetricsreceiver/ — all disk/plausibility tests pass

Requirements

1. Run make fmt and make fmt-sh
2. Run make lint