feat(pruning): PruningConjunction with per-leaf observation

Adds `PruningConjunction` — an AND of tagged `PruningPredicate` leaves —
plus a `PruningObserver` trait fired once per leaf actually evaluated.
The AND short-circuits once every container is pruned; short-circuited
leaves are not observed, so per-conjunct stats are never biased by
predicates that did not run.

Construction is via a small builder:

    let conj = PruningConjunction::builder(schema)
        .push(filter_id, expr)   // runs try_new, skips always-true / errors
        .build();                // Option<PruningConjunction>

`PruningConjunction::single(predicate)` wraps one already-built
predicate for the untagged path, which uses `prune()` / `NoopObserver`
at zero overhead.

`ConjunctStatsObserver` accumulates `PerConjunctPruneStats { tag,
containers_seen, containers_pruned }`.

OR and NOT are intentionally *not* structural nodes: they are handled
inside a leaf via `PruningPredicate::try_new`, which both prunes better
(cross-branch reasoning) and keeps the per-leaf "containers pruned"
count sound (it is only monotonic under AND). Per-branch OR
short-circuit, where it is useful, belongs to the row-filter decode
path over exact masks — not here.

Replaces the placeholder-`true`-literal `sub_predicates` design and its
double traversal from the earlier draft (#22235).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: adriangb

datafusion/pruning/src/lib.rs

@@ -18,9 +18,14 @@
 #![cfg_attr(test, allow(clippy::needless_pass_by_value))]
 
 mod file_pruner;
+mod pruning_conjunction;
 mod pruning_predicate;
 
 pub use file_pruner::FilePruner;
+pub use pruning_conjunction::{
+    ConjunctStatsObserver, NoopObserver, PerConjunctPruneStats, PruningConjunction,
+    PruningConjunctionBuilder, PruningObserver, Tag,
+};
 pub use pruning_predicate::{
     PredicateRewriter, PruningPredicate, PruningStatistics, RequiredColumns,
     UnhandledPredicateHook, build_pruning_predicate,