prometheus_remote_write source: skip_nan_values does not cover histogram/summary counts, stale-marker NaN still 400s the whole request

skip_nan_values: true (added in #23774) filters NaN f64 sample values, but the histogram/summary count conversion still rejects the whole write request when a count sample is NaN:

WARN source{component_id=prometheus_remote_write_source component_type=prometheus_remote_write}:
  Received bad request. error=Could not decode write request:
  expected value in range [0, 18446744073709551615], found: NaN
  error_code=http_response_400 http_code=400

Observed on vector 0.57.0 with skip_nan_values: true active, receiving from VictoriaMetrics vmagent.

The path is try_f64_to_u64() in lib/prometheus-parser/src/lib.rs, used for histogram bucket / summary counts during aggregation. It is unconditional; skip_nan_values never reaches it.

NaN counts are not malformed input: the remote-write 1.0 spec requires senders to mark stale series with a NaN value, so any scrape target disappearing can put NaN on a _count/_bucket series. Senders such as vmagent treat the 400 as non-retryable and drop the entire block, so every unrelated sample batched alongside is silently lost.

Expected: with skip_nan_values: true, NaN counts (at minimum stale markers) are skipped like other NaN samples instead of failing the request.

Start in lib/prometheus-parser/src/lib.rs at try_f64_to_u64(), then trace its histogram bucket and summary count callers to see where skip_nan_values is available. Use the linked remote-write stale-marker specification as the behavior reference; done means NaN counts are skipped when skip_nan_values is enabled instead of rejecting the whole request.