RFC 0059: reframe as an FSST-family encoding, amplify OnPair credit

Reposition the encoding as descended from FSST — a cousin Array to
FSSTArray in the same encodings/fsst crate, following the
vortex-fastlanes precedent (one family-named crate, multiple
algorithm-named Array types). The substantive algorithmic innovation
the new encoding carries is OnPair's; the architectural archetype
is FSST's. Both deserve credit, in proportion to their respective
contributions.

Concretely:

- Reframe the Summary to lead with "adds a new array to Vortex's
  existing FSST encoding family — a cousin to FSSTArray". The four
  contributing prior works are listed in order of how much of the
  new encoding is theirs: OnPair (the codebook construction + the
  OnPair16 decode pattern + the on-disk dictionary representation +
  the compressed-domain automata), FSST (the family archetype),
  FSST12 (the bit-packing layout at 12-bit width), GSST (the GPU
  decoder shape).

- Restructure Motivation to start with FSST as the field-level
  random-access archetype, then frame OnPair / FSST12 / GSST as
  improvements within the FSST family rather than alternative
  designs. Add an explicit note that OnPair "has received less
  attention than its contributions warrant" and that this RFC
  credits accordingly.

- Rewrite Lineage with the family-tree framing: FSST is the
  archetype, OnPair is the algorithmic improvement we're adopting,
  FSST12 contributes the bit-packing, GSST contributes the GPU
  shape. Add an explicit note pointing readers to the OnPair paper
  as the primary source.

- Update Current Vortex state to position the new Array as a cousin
  in the same crate, with an analogy to BitPackedArray /
  DeltaArray / FoRArray cohabiting encodings/fastlanes. Add a link
  to encodings/fastlanes as the crate-organization precedent.

- Restructure "Why this is the right design" to lead with "why a
  cousin Array rather than a flag on FSSTArray" — the family
  framing changes the natural objection.

- Restructure Prior Art with OnPair listed first and given a full
  prose treatment that names Gargiulo and Venturini, summarizes the
  paper's four substantive contributions, and explicitly notes
  recognition gap. FSST appears second as the family ancestor.
  Vortex integration templates are listed separately, with
  encodings/fastlanes added as the crate-organization precedent.

- Update Compatibility to emphasize "two new types in the existing
  FSST encoding crate, not a change to existing arrays."

- A note on naming added to the Summary: the type name is
  OnPair16Array rather than something with "FSST" in it because the
  substantive algorithmic contribution is OnPair's; the crate
  location (encodings/fsst) handles the family signaling, matching
  the FastLanes precedent.

Net diff: 546 lines (was 490), 167 insertions, 111 deletions.
The technical content is unchanged; the framing makes the lineage
and credit attribution match the actual contributions.

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

Author: lwwmanning