MQE: Eliminate deduplicate and merge nodes for binary expressions

[56quarters]

What this PR does

This change extends the EliminateDeduplicateAndMergeOptimizationPass to attempt to remove deduplicate and merge nodes on either side of a binary expression.

Which issue(s) this PR fixes or relates to

Prerequisite for #13863

Checklist

• Tests updated.
• Documentation added.
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]. If changelog entry is not needed, please add the changelog-not-needed label to the PR.
• about-versioning.md updated with experimental features.

---

Note

Improves query plan optimization by pruning unnecessary DeduplicateAndMerge nodes around binary expressions, respecting delayed name removal semantics.

• Enhances EliminateDeduplicateAndMergeOptimizationPass to keep only the required dedupe node for binary ops and eliminate others when safe
• Adds Prometheus counters: cortex_mimir_query_engine_eliminate_dedupe_attempted_total and cortex_mimir_query_engine_eliminate_dedupe_modified_total; increments on attempts/plan changes
• Planner now registers the optimization pass with a Registerer; all call sites updated
• Test suite updated to reflect simplified plans (fewer dedupe nodes), assert new metrics, and adjust dispatcher node paths/indices
• Minor cleanup: simplify types.HasDuplicateSeries for 2-series case

^{Written by Cursor Bugbot for commit ae195fe. This will update automatically on new commits. Configure here.}

[charleskorn]

Overall LGTM. Could you please add a test for cases where a binary operation is nested? (eg. foo or bar or baz, or rate(foo[1m]) or rate(bar[1m]) or rate(baz[1m]))

[charleskorn]

LGTM modulo suggestions below

[cursor]

Label replace handling incorrectly marks sibling branch nodes

Medium Severity

The label_replace handling at lines 148-150 assumes nodes[len-2] is always an ancestor of the label_replace function. However, with the new binary expression traversal (lines 93-103), the nodes list can now contain DeduplicateAndMerge nodes from sibling branches. For a query like rate(foo[5m]) or label_replace(rate(bar[5m]), ...), when processing label_replace, nodes[len-2] incorrectly points to foo_rate's dedup node (from the LHS branch) rather than an ancestor of label_replace. This causes the LHS dedup node to be incorrectly retained when it should be eliminated (since foo has an exact name matcher). Before this PR, binary expressions caused early return so this code path was never exercised.

 

[56quarters]

This results in keeping an extra DeduplicateAndMerge node which is less efficient than dropping it but doesn't cause in incorrect results. I'll handle this in a follow up PR.