Store correctness contract test suite

[zzylol]

Summary

Adds a store_correctness_tests module that defines a contract test suite — a single run_contract_suite function called from two #[test] entry points — one for each LockStrategy. Any future Store implementation can be validated by calling run_contract_suite with its factory.

Coverage (33 sub-tests per strategy = 66 assertions total)

Empty-store edge cases

• Range query on empty store returns empty map
• Exact query on empty store returns None
• Earliest timestamp on empty store returns empty map

Single insert

• Range query within bounds returns the entry
• Range query outside bounds returns empty
• Exact query hit (matching start and end)
• Exact query with wrong start returns None
• Exact query with wrong end returns None

Batch inserts

• Full-range query returns all inserted entries
• Results are chronologically ordered (ascending by window start)

Partial-range filtering

• Only windows overlapping the query range are returned

Aggregation-ID isolation

• Querying one agg-ID does not return entries from another

Earliest-timestamp tracking

• Minimum timestamp is tracked across multiple inserts
• Each aggregation ID tracks its own minimum independently

Cleanup policies

• CircularBuffer: evicts oldest windows when count exceeds threshold
• CircularBuffer: newest windows are retained after eviction
• ReadBased: evicts windows after the configured read-count threshold
• ReadBased: windows below the threshold are retained

DeltaSetAggregator exclusion

• DeltaSetAggregator-typed agg IDs bypass cleanup entirely, even when the circular-buffer policy would otherwise evict

Keyed (label-grouped) entries

• Entries with a KeyByLabelValues key are grouped correctly in query results
• Keyed and unkeyed entries for the same agg-ID coexist without interference
• Multiple distinct keys in the same time window are returned under separate map entries

Clone fidelity (all 11 accumulator types)

Each test inserts an accumulator into the store, queries it back, calls clone_boxed_core(), and asserts serialize_to_json() output is identical:

Accumulator	Test
SumAccumulator	non-zero sum preserved
MinMaxAccumulator	min/max values preserved
DatasketchesKLLAccumulator (k=200)	sketch with 20 data points preserved
IncreaseAccumulator	value preserved
MultipleSumAccumulator	multi-key sums preserved
MultipleMinMaxAccumulator	multi-key min/max preserved
SetAggregatorAccumulator	set of label-value keys preserved
DeltaSetAggregatorAccumulator	added/removed key sets preserved
CountMinSketchAccumulator	initialised sketch preserved
CountMinSketchWithHeapAccumulator	initialised sketch with heap preserved
HydraKllSketchAccumulator	multi-shard sketch with data preserved

Concurrency

• 8 threads inserting concurrently produce no data loss (all windows queryable)
• Concurrent reads from multiple threads all return complete results

Test plan

• cargo test -p query_engine_rust store_correctness — 2 tests pass (PerKey + Global), 66 sub-assertions each
• cargo fmt --all — no formatting issues

🤖 Generated with Claude Code