Ballyhoo Bloom in F*#

The Bloom sidecar story is no longer only Python/runtime glue.

New module:

• Parser.BallyhooBloom.fst

Current scope:

• F* representation of predicate Bloom filters
• deterministic hash/position generation
• insert
• membership test
• OR-rollup/union for compatible filters

This is intentionally smaller than a full sidecar file parser. The immediate goal is to make the in-memory Bloom structures and utility logic ours, in F*, so they can be used alongside HDT-backed stores and later in purely in-memory corpus/index experiments.

Short-term split:

• Python still generates the current graph.bloom.pred.json and graph.bloom.pred.bin sidecars
• OCaml runtime glue still consumes those sidecars on the HDT path
• F* now owns the Bloom data structure model and core operations

Next obvious extensions:

• fixed-size sidecar byte representation in F*
• sidecar parsing/serialization
• subset/bundle Bloom summaries under corpus TOC areas
• use the same structure for in-memory ephemeral corpus pruning, not only on-disk HDT artifacts