Fix #3953: Dict Layout to short-circuit fetching codes for list_contains

[pamod-madubashana]

Fixes #3953

What changed

DictReader::pruning_evaluation in vortex-layout/src/layouts/dict/reader.rs now handles list_contains(root, needle) filters.

When the needle side of the expression does not reference the root (typically a literal), the predicate is evaluated against just the small dictionary values array via the existing values_eval cache. If none of the dictionary values contain the needle (nulls treated as false, matching filter semantics), pruning returns an all-false mask for the full row range. Otherwise it returns the input mask unchanged.

Because pruning never touches the codes child, a fully pruned range short-circuits the scan before filter_evaluation ever fetches codes — exactly the expensive IO this avoids. Non-list_contains expressions keep the previous behavior (no values read during pruning), and list_contains expressions that do reference the root on the needle side are also left alone since they cannot be evaluated on the values array alone.

Why

A dict-encoded list column can answer list_contains membership from its dictionary alone when the answer is negative: if no dictionary entry contains the needle, no row can match. Previously pruning was a no-op for the dict layout, so every list_contains filter paid for a full codes read even when the dictionaries already proved the result empty.

Verification

Attempted cargo clippy -p vortex-layout --all-targets -- -D warnings, but it cannot build in this environment: the vortex-array build script requires the FlatBuffers compiler (failed to run flatc: No such file or directory). No test could be run as a result; the change was verified by careful reading against the existing filter_evaluation values-eval/mask logic instead.

[robert3005]

The reason we haven't done this kind of operations is that we shouldn't select on expression id. There has to be a mechanism that makes it generic

[pamod-madubashana]

Addressed: replaced the ListContains-specific branch with a generic rule. pruning_evaluation now handles any boolean filter that references the root and whose whole tree is infallible: it evaluates the filter against the cached dictionary values and prunes the range when all values miss (nulls as false, matching filter semantics). This covers list_contains as well as eq/like/etc. without matching on expression id, and reuses the values_eval cache so a non-pruned range pays no second values read.