refactor(prometheus): simplify _CustomCollector by replacing deque with Optional[MetricsData]

[sridhar-3009]

Summary

Fixes #2500 (referenced in PR #2321 discussion).

The internal _CustomCollector used a deque[MetricsData] to buffer metric data between the _receive_metrics → add_metrics_data call and the Prometheus collect() scrape. In practice the deque holds at most one item per scrape cycle because _callback() triggers exactly one _receive_metrics invocation before collect() drains the queue.

This PR simplifies the flow:

• Replace deque[MetricsData] with MetricsData | None
• add_metrics_data assigns directly instead of appending
• collect() reads and clears the single field, replacing the while … popleft() loop
• Remove the now-unused from collections import deque import

This also fixes a latent correctness issue: with the deque, if add_metrics_data were called multiple times between scrapes, collect() would yield duplicate metric family entries (because metric_family_id_metric_family accumulated across the loop but was yielded inside the loop).

Before

collect() → callback() → _receive_metrics() → add_metrics_data() → deque.append()
         → while deque: popleft() → translate → yield (inside loop, potential duplicates)

After

collect() → callback() → _receive_metrics() → add_metrics_data() → self._metrics_data = …
         → translate single MetricsData → yield once

Test plan

• Existing Prometheus exporter tests pass
• Manual scrape endpoint returns correct metrics
• No metrics emitted when no data collected (metrics_data is None)

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: sridhar-3009 / name: Sai Sridhar (db88905)