fix(metrics): restore IndexMap for unbounded MetricSet to fix CPU regression

[thomasqueirozb]

Summary

MetricSet was changed from IndexMap to LruCache in #23374 to support the incremental_to_absolute transform's capacity eviction feature. However, LruCache::get_mut unconditionally updates LRU bookkeeping (pointer manipulation in a doubly-linked list) on every metric lookup, even when no capacity policy is configured. For high-throughput metrics pipelines (e.g. prometheus_remote_write sinks processing thousands of metrics/second) this per-metric overhead accumulates significantly on the hot path and is the root cause of the CPU regression reported in #24058 that persists in v0.55.0 after the buffer hot-path fix (#24995) and file source fix (#25064).

MetricSet now uses an either IndexMap or LruCache if needed, meaning that performance should decrease unless the user is using the incremental_to_absolute transform.

Vector configuration

NA

How did you test this PR?

NA

Change Type

• Bug fix
• New feature
• Dependencies
• Non-functional (chore, refactoring, docs)
• Performance

Is this a breaking change?

• Yes
• No

Does this PR include user facing changes?

• Yes. Please add a changelog fragment based on our guidelines.
• No. A maintainer will apply the no-changelog label to this PR.

References

• Potential fix for: Vector CPU Usage increased 0.50.0 #24058
• Related: feat(new transform): Add new incremental_to_absolute transform and change to MetricSet to an LRU cache with optional capacity policies #23374