[METRICS] Support synchronous gauge with delta temporality

[pranitaurlam]

Fixes #4146

Summary

Removes the block in MetricCollector::GetAggregationTemporality that logged an error and silently fell back to cumulative when delta was requested for a synchronous gauge instrument.

The existing delta fast-path in TemporalMetricStorage::buildMetrics already provides the correct window-scoped last-value semantics for gauge: because SyncMetricStorage::Collect swaps and resets the live attributes_hashmap_ on every collection, delta_metrics contains only values recorded since the previous collection. Stale attribute sets that were not updated in the current window are simply absent from the snapshot and are never re-emitted.

Changes:

• metric_collector.cc: remove the gauge+delta error/fallback; return the reader-requested temporality directly.
• temporal_metric_storage.cc: update fast-path comment to mention gauge and its window-scoped last-value behavior.
• sync_metric_storage_gauge_test.cc:
  • Add kDelta to both INSTANTIATE_TEST_SUITE_P suites and assert values for both temporalities.
  • Add GaugeDeltaWindowTest.NoRecordWindowEmitsNothing: record → collect → no record → collect emits nothing.
  • Add GaugeDeltaWindowTest.StaleAttributeNotReemitted: record A+B → collect → record A only → collect emits only A.
  • Add GaugeDeltaWindowTest.RecordAfterEmptyWindowEmitsNewValue: empty window followed by a new recording is exported correctly.

Test plan

• Existing cumulative gauge tests still pass with both temporalities.
• NoRecordWindowEmitsNothing: second collection with no new recordings emits 0 points.
• StaleAttributeNotReemitted: attribute not recorded in the current window is not re-exported.
• RecordAfterEmptyWindowEmitsNewValue: new recording after an empty window is exported correctly.

[pranitaurlam]

Hi @lalitb @marcalff @ThomsonTan — I've opened #4152 to address this.

The fix removes the error/fallback in MetricCollector::GetAggregationTemporality and relies on the existing delta fast-path in TemporalMetricStorage, which already provides the correct window-scoped last-value semantics (the live attributes_hashmap_ is swapped and reset on every collection, so stale attribute sets are never in the snapshot).

Tests added for the scenarios you mentioned: record → collect → no record → collect (emits nothing), stale attribute not re-emitted, and recovery after an empty window. Happy to adjust based on your feedback.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.09%. Comparing base (4352a63) to head (66744a6).
⚠️ Report is 8 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4152      +/-   ##
==========================================
+ Coverage   82.01%   82.09%   +0.08%     
==========================================
  Files         385      386       +1     
  Lines       16093    16207     +114     
==========================================
+ Hits        13197    13303     +106     
- Misses       2896     2904       +8     

Files with missing lines	Coverage Δ
sdk/src/metrics/state/metric_collector.cc	96.62% <100.00%> (+2.96%)	⬆️
sdk/src/metrics/state/temporal_metric_storage.cc	97.78% <ø> (+0.06%)	⬆️

... and 29 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[lalitb]

Can we also add a multi-reader case here? These tests cover the single-reader delta path, but with multiple readers the storage uses the shared unreported_metrics_ path. It would be good to verify record -> collect -> no record -> collect for a delta gauge there too, so stale gauge values are not emitted again.

[niqiangpro-cell]

I added tests for multi reader case in this PR #4159. It verifies the expected bahaviour for multi reader path as well. What I understood is that instead of merging delta_metrics with last reported metrics like what is done for cumulative temporality, we only need to export delta_metrics as is, i.e., not emit stale values.

[lalitb]

@pranitaurlam - Since #4159 already carries the production-code change and adds the multi-reader coverage, can we narrow this PR to only the single-reader delta gauge tests that are not covered there? That would mean removing the MetricCollector::GetAggregationTemporality() change from this PR and keeping only the additional single-reader coverage, so the implementation change lands in one place.