refactor(metrics): replace swap-and-drain delta collection with in-place iteration

[bryantbiggs]

Splits from #3392, per maintainer feedback. Refs #2328.

Summary

• Replaces the two-HashMap swap-and-drain pattern in collect_and_reset with in-place iteration using TrackerEntry status tracking
• TrackerEntry wraps each aggregator with has_been_updated (AtomicBool) to track whether measurements occurred since the last collection cycle
• Removes has_no_attribute_value — the no_attribute_tracker.has_been_updated flag now serves the same purpose
• collect_and_reset now iterates under a read lock (no write lock in steady state), exports only updated entries, and evicts stale entries under a write lock with TOCTOU re-check
• Adds drain_and_reset for Observable/async instruments that need true staleness detection via map clearing
• Eliminates O(n) write-lock acquisitions on the hot path per collection cycle when attribute sets are reused (the common case)

User-facing behavior change

Recovery from cardinality overflow now requires 2 collect cycles instead of 1 — the first cycle marks entries as stale (no updates), the second cycle evicts them. This trade-off enables the in-place iteration pattern that avoids write-lock contention in steady state.

Locking trade-offs (discussed with @cijothomas and @utpilla in #3392)

Scenario	Old (swap-and-drain)	New (in-place)
Steady state (reused attrs)	O(n) write locks per cycle	Read locks only
New/returning attrs during collect	Write lock for insert	Write lock for insert (competes with read lock held by collect)
Stale eviction	Implicit (map cleared)	Write lock proportional to stale count

Test plan

• All 149 existing metrics tests pass
• counter_aggregation_overflow_delta tests updated for two-phase eviction semantics
• No changes to cumulative collection path (collect_readonly)
• cargo fmt --all -- --check passes
• cargo clippy -p opentelemetry_sdk --features metrics -- -D warnings passes

[codecov]

Codecov Report

❌ Patch coverage is 94.85294% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.3%. Comparing base (8e95e16) to head (35f7e52).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
opentelemetry-sdk/src/metrics/internal/mod.rs	93.9%	7 Missing ⚠️

Additional details and impacted files

@@          Coverage Diff          @@
##            main   #3420   +/-   ##
=====================================
  Coverage   83.2%   83.3%           
=====================================
  Files        128     128           
  Lines      25086   25177   +91     
=====================================
+ Hits       20896   20978   +82     
- Misses      4190    4199    +9     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[bryantbiggs]

CI integration test failure (counter_non_tokio) is pre-existing flaky test — UUID collision from parallel tests sharing the collector. Previous run on this branch passed. Our changes only touch opentelemetry-sdk/src/metrics/internal/mod.rs.

[bryantbiggs]

Opened #3424 to fix the flaky integration test that showed up in CI here.

[cijothomas]

Thanks!