Fix ingester panic on float histogram with zero count

[yeya24]

What this PR does

The ingester Push path selected the float vs integer native-histogram decoder using a value check, hp.GetCountFloat() > 0:

if hp.GetCountFloat() > 0 {
    fh = cortexpb.FloatHistogramProtoToFloatHistogram(hp.Histogram)
} else {
    h = cortexpb.HistogramProtoToHistogram(hp.Histogram)
}

A float histogram with a count of 0 (for example a staleness marker — FloatHistogram{Sum: StaleNaN} — or an empty histogram) still has the CountFloat oneof set, so IsFloatHistogram() reports true while GetCountFloat() > 0 is false. Such samples were routed to HistogramProtoToHistogram, which panics:

panic: HistogramProtoToHistogram called with a float histogram

Because this runs synchronously in the gRPC Push handler with no recover, the panic crashes the whole ingester process. With many ingesters receiving the same input this can break write/read quorum across a cell.

Fix

Select the decoder by the proto type via IsFloatHistogram(), matching the discriminator already used in pkg/util/validation and upstream Prometheus.

Testing

Added TestIngester_Push_FloatHistogramWithZeroCount covering a staleness-marker float histogram and an empty float histogram (both count 0). Verified the test panics on master without the fix and passes with it. Full pkg/ingester suite passes with -tags "netgo slicelabels".

Which issue(s) this PR fixes

N/A

Checklist

• Tests updated
• CHANGELOG.md updated - (will add with PR number)
• docs/configuration/config-file-reference.md updated (regenerated; no change — no config flags added)

[SungJin1212]

lgtm