feat(executor): exact distinct, L2 and entropy, and typed UnivMon identities

[zzylol]

Rebased on main d4869a7 (DF 54).

Wave 1 chain: #594 → #593 → #592 → #591 → #595 → #596 → #597 → #598 → #599 (on #589)

Why

#509 Example 2 plans one UnivMon over src_ip for COUNT(DISTINCT src_ip), the entropy of the src_ip distribution and the L2 norm of per-src_ip counts. Pass 1 also keeps an exact candidate for each statistic. On the current stack none of these runs natively:

• the executor has no native UnivMon state or readouts (summary family has no native DAG state implementation), and a summary build accepts only Float64 updates;
• Cardinality, FrequencyL2 and FrequencyEntropy aggregates fail to bind (aggregate intent has no native implementation).

The parked draft PRs #557, #559 and #563 fixed this on the old stack (asap-physical-operators). This PR ports them onto crates/executor. Part of #509 / #580 (W8).

What

Before	After
Operator::summary_build(flows(src_ip: Utf8), UnivMon, ..) → Err("summary numeric update requires Float64")	one UnivMon state from Utf8/Int64/Bool identities; Cardinality, L2 and entropy readouts all read it
Aggregate [Cardinality / FrequencyL2 / FrequencyEntropy {accuracy: Exact}] → no native implementation	exact reducers: distinct tuples, L2 of unit counts, entropy in bits
SELECT COUNT(DISTINCT src_ip) FROM flows WHERE keep cannot execute	executes through a raw connector: 0 / 0 / 2 on the test inputs

How

• UnivMon state (the old stack's 14c8ac8, a prerequisite feat(runtime): accept typed UnivMon frequency identities #557 assumed). validate_native_family, validate_sketch_evaluation and validate_state accept UnivMon with bare count, Cardinality, FrequencyL2 and FrequencyEntropy readouts.
• feat(runtime): accept typed UnivMon frequency identities #557 typed identities. AccumulatorUpdater::update_value(&Value, ts) defaults to the Float64 contract. UnivMon inserts Bool/Int64/Utf8 as type-tagged Value::key() bytes, so Int64 keys above 2^53 stay distinct. Heap key bytes count toward state memory.
• feat(runtime): execute exact frequency L2 and entropy #559 exact L2 and entropy. Reduction::FrequencyL2/FrequencyEntropy(col): NULL skipped, non-finite floats rejected, 0.0 for an empty population, entropy in bits, per-key memory reservation and cooperative checkpoints.
• feat(runtime): execute exact distinct identity counts #563 exact distinct. Reduction::Cardinality(cols): a tuple with any NULL is skipped; the grouped key encoding keeps tuple boundaries and folds signed zero and NaN payloads.
• The legacy raw-cost adapter (plan-selection/src/cost/query_physical_lowering.rs) lists the three intents as hash aggregates, as feat(runtime): execute exact frequency L2 and entropy #559 and feat(runtime): execute exact distinct identity counts #563 did. plan_stages already prices any aggregate.

Name and path changes only: asap_physical_operators → asap_executor, pre_asap/post_asap → types::ir::{operator, schema, properties}, PostAsapDAGNode → PhysicalASAPDAGNode. The test helper execute_raw_rows (from #563) uses compile_physical_asap_dag and roots[0]. The developer-doc notes from #557 and #559 are not ported.

Tests

Carried over from #557 (univmon_execution::typed_frequency_keys_preserve_identity, two kernel tests), #559 (exact_frequency_*, the memory-budget test) and #563 (exact_cardinality_*, sql_cardinality), plus 14c8ac8's one_univmon_state_answers_distinct_l2_and_entropy. precompute_raw_samples now expects UnivMon to compile and checks its three readouts (the old stack's d206f44), because UnivMon has native state. Before the port, both UnivMon tests fail at runtime and the reducer tests do not compile (no variant FrequencyL2/Cardinality).

Gate: cargo fmt --all --check and cargo clippy --workspace --all-targets --all-features -- -D warnings pass. cargo test --workspace: 1,527 passed, 10 ignored (#589: 1,517 / 10).

Ports #557, #559, #563. Next: SQL recognition (#598).

🤖 Generated with Claude Code