feat(ingester): Add head-only queried series metrics

[yeya24]

What this PR does

This PR adds two new metrics that track query activity against the TSDB head only (in-memory, ~2h data):

1. cortex_ingester_queried_series_from_head — Estimated unique series queried from head within a configurable time window (HLL-based, deduplicated across queries).

2. cortex_ingester_queried_metric_series_in_head — Current head cardinality for each metric name that was queried within the configured window.

Motivation

The existing cortex_ingester_active_queried_series metric tracks all queried series regardless of whether they come from head or on-disk blocks. For understanding active query patterns and resource usage, we specifically need to know what's being queried from the head (the hot, recent data).

Implementation

• Uses a BlockChunkQuerierFunc wrapper that intercepts Select calls only for head queriers (identified by RangeHead/Head ULIDs)
• Reuses existing ActiveQueriedSeries HLL and ActiveQueriedSeriesService for the total series metric
• Per-metric-name tracking uses a simple map[string]time.Time with lazy count lookup via seriesInMetric.getSeriesCountForMetric() on periodic collection
• Only __name__= equality matchers trigger per-metric tracking
• Sampling supported for the HLL metric to reduce overhead
• Both metrics share the same configurable time window

New configuration flags (experimental)

Flag	Default	Description
-ingester.head-queried-series-metrics-enabled	false	Enable the feature
-ingester.head-queried-series-metrics-windows	2h	Time windows to report
-ingester.head-queried-series-metrics-window-duration	15m	HLL sub-window bucket size
-ingester.head-queried-series-metrics-sample-rate	1.0	Query sampling rate

Checklist

• Tests updated
• Documentation added
• v1-guarantees.md updated (experimental feature)