There are already methods operating on ByteBuffer, however versions
taking streams allow to avoid data copying, and more directly compatible
with various network APIs. Also, for reference, Guava Bloom filter implementation
uses streams.

There's a bit of code duplication as the result. Sadly, there's no standard
class to wrap ByteBuffer in a stream, and pulling in an external dependency
seems like an overkill here.

Xor: implement stream serialization

Implements the standard (non-homogeneous, w=64) Ribbon filter
  construction and membership query described in issue FastFilter#33, following
  Xor8's structure (hash-derive, build-with-retries, query-by-recombine).

Mirrors Xor8's ByteBuffer/OutputStream serialize + static deserialize
  convention. Unlike Xor8, numSlots/numStarts must be stored explicitly
  (not re-derived from size) since the construction retry loop can grow
  the slot count beyond the simple sizing formula.

Setting is ignored the same way XOR_8/XOR_16/XOR_PLUS_8 ignore it here:
  TestAllFilters.test() hardcodes setting=10 for every TestFilterType,
  which would exceed RibbonFilter's 8-bit (byte-storage) ceiling if
  passed through.

Reserves the RibbonFilter/RIBBON name for a future
  ICML-backed implementation with fast queries; this one keeps
  the naive O(w*r) per-column query, which is 70-100x slower
  than the Xor filters (see PR FastFilter#54 benchmark).

Add Ribbon filter implementation