feat: add exemplar support to timeseries responses for individual profile retrieval

[marcsanmi]

This PR implements backend support for exemplars in timeseries queries.

Changes

Protobuf API

• Add Exemplar message with profile_id, timestamp, value, and labels fields
• Extend Point message with exemplars field to embed exemplar metadata in timeseries responses
• Add ExemplarType enum to control exemplar inclusion (NONE, INDIVIDUAL)

Query Pipeline

• Thread profileID through v2 query backend via ProfileEntry.ID
• Read profile UUID from Parquet ID column during timeseries queries
• Use withFetchProfileIDs(true) option to enable profile ID fetching only when exemplars are requested
• Maintain v1 compatibility by passing empty profileID in legacy query paths

Exemplar Management

• ExemplarBuilder: New dedicated component for per-series exemplar tracking
  • Deduplicates exemplars by fingerprint (label set hash)
  • Merges exemplars with same (profileID, timestamp) by intersecting labels
  • Handles dynamic labels (e.g., pod changes during migration)
• Early label filtering: Filter out groupBy labels at ingestion time using efficient WithoutLabels() method
• Three-level label intersection:
  a. ExemplarBuilder: Within single block
  b. TimeSeriesMerger: Across blocks at same timestamp
  c. RangeSeries: Across timestamps within time bucket

Merging & Aggregation

• Single-block path: Exemplars attached to points with matching timestamps
• Multi-block path: TimeSeriesMerger merges exemplars across blocks, preserving highest values
• sumTimeSeriesAggregator and avgTimeSeriesAggregator preserve and merge exemplars during re-aggregation
• Top-N selection per point (default: 1, configurable via maxExemplarsPerPoint)

Performance Optimizations

• Efficient sorted-list attachment algorithm (O(E+P) instead of O(E*P))
• Conditional ID column fetching only when exemplars requested
• In-place label filtering using Subtract() method
• Early filtering eliminates redundant aggregator-level filtering

Implementation Flow

1. v2 single-block: queryTimeSeries → TimeSeriesBuilder.Add() → filter labels → ExemplarBuilder.Add() → BuildWithExemplars()
2. v2 multi-block: TimeSeriesBuilder → TimeSeriesMerger.MergeTimeSeries() → RangeSeries() re-aggregates with exemplars
3. v1 paths: Empty profileID for backward compatibility (no exemplars)

[simonswine]

Just focused on the query-backend for now

[simonswine]

LGTM

Great work, thanks for going through all my comments