Review-driven refinements to the query no-index-lookup counter:

- Count per-query, not per-operator: a query with N qualifying ScanAll+Filter
  scans now bumps memgraph_query_no_index_lookup_total by 1. Makes
  rate(no_index)/rate(queries) a stable alert ratio. The per-operator count is
  still produced in PlanHintsResult for future structured use.
- Unit-test the counter: query_hint_provider asserts no_index_lookup_count in
  every case (Cartesian -> 2, ScanAllByLabel+property -> 0). Was untested before.
- Rename PlanHintsProvider::hints() -> take_hints() (returns by value) to drop
  the misleading std::move on an lvalue-ref accessor.
- Reword the counter Help string for per-query semantics; document the
  intentional ScanAllByLabel non-increment and the OpenMetrics/SHOW-only scope
  (no deprecated JSON-global aggregation).
- e2e: UNION asserts +1 (per-query); add a PROFILE no-op assertion.

The exposed metric is per-query (counts once when a plan has any qualifying
ScanAll+Filter), so the internal per-operator count was never used as a
magnitude. Replace PlanHintsResult::no_index_lookup_count (size_t) with
has_no_index_lookup (bool): expresses the actual contract, removes a field
that had to stay in sync with hints_, and makes per-operator exposure
impossible to reintroduce by accident. Branch-firing coverage is retained via
the existing hint-message assertions.

"no_index_lookup" was ambiguous (a lookup that found no index, or the absence
of a lookup?) and didn't say a scan happened. Rename to the clearer, convention-
aligned (memgraph_*_queries_total) name before release:

- metric: memgraph_query_no_index_lookup_total -> memgraph_unindexed_scan_queries_total
- SHOW METRICS INFO row: QueryNoIndexLookup -> UnindexedScanQueries
- family/handle: query_no_index_lookup* -> unindexed_scan_queries*
- PlanHintsResult field: has_no_index_lookup -> has_unindexed_scan

Pure rename; behavior and tests unchanged.

…ning

The plan-hint provider classified scans with an exact `GetTypeInfo() ==
ScanAll::kType` check. Parallel-execution plans rewrite `Filter -> Scan*`
into `Filter -> ScanChunk -> ParallelMerge -> ScanParallel*`, so the filter
sits on a `ScanChunk` (its own kType), which fell through silently and never
incremented `memgraph_unindexed_scan_queries_total`.

Add `EffectiveScanType` to resolve a `ScanChunk` down to the underlying
`ScanParallel*` and map it back to the equivalent sequential type
(`ScanParallel` -> `ScanAll`, `ScanParallelByLabel` -> `ScanAllByLabel`), so
parallel plans count identically to sequential ones. Index-backed parallel
scans stay unflagged.

Review hardening:
- reuse the guarded `dynamic_cast<ScanAll*>` instead of casting twice
- make `take_hints()`/`has_unindexed_scan()` private + `friend ProvidePlanHints`
  (+ `[[nodiscard]]`) so result accessors can't be read mid-traversal
- extract `EmitPlanHints` for the three duplicated hint loops; keep the counter
  increment only on the regular-execution path, with explicit comments on the
  EXPLAIN/PROFILE no-op and the before-AccessorCompliance ordering
- `Increment(1.0)` -> `Increment()`
- document the property-only-filter non-count case and the JSON-endpoint omission

Tests:
- unit: parallel unindexed scan counts, parallel label scan does not,
  property-only sequential scan does not
- e2e: `USING PARALLEL EXECUTION` unindexed scan counts (+1) and is silenced by
  a matching index

Apply consolidated review feedback (no behavior change to the metric):
- Fix misleading comment: ScanParallelByEdge is a full edge scan, not
  index-backed; state the out-of-scope reason explicitly.
- Reword counter help text to "Planned queries" to match counting before
  AccessorCompliance.
- Mark ProvidePlanHints [[nodiscard]].
- Move HintIndexUsage and EffectiveScanType out-of-line into the .cpp.
- Align metric section labels with the emitted "Query" category.
- Add regression tests: property-only parallel scan and ScanChunkByEdge
  parallel pass-through; make WrapInParallelChunk a fixture helper.
- Condense over-long comments.

- assert ScanChunk/ParallelMerge chain invariants instead of dead null guards
- add unit test covering the ScanParallelByLabelProperties (indexed) branch
- spell out trigger and exclusion conditions in the counter help text
- trim verbose comments to one line; fix the misleading edge-scan note

Replace the no-op DMG_ASSERTs in EffectiveScanType (compiled out under
NDEBUG) with always-on guards that fall back to the scan's own type when
the parallel-rewrite chain has an unexpected shape, so a malformed chain
can no longer misclassify or null-deref in release builds.

Also tighten review-flagged comments: drop the chain description that was
duplicated from the function header, correct the EmitPlanHints contract
comment (it never touches the counter), and collapse the metric-handle
comment to one line.

Reword the metric Help and the EffectiveScanType / has_unindexed_scan
comments so they cover parallel-execution plans (not just the sequential
ScanAll), correct the parallel-rewrite chain description, and fix a
dropped word. Comment/string only; no behavior change.