RFC 0059: move staging from Design to Implementation Specification

Per author feedback: "Stages are often more of an implementation detail
than part of the actual design." Removing the Staging subsection from
the Design narrative; recasting as a "Suggested rollout plan" inside
the Implementation Specification section (where it belongs as a
project-plan note, not a design choice).

Changes:

- Drop the "Staging" subsection from Design entirely (was between
  "Hidden assumptions" and "Diagrams").
- Drop the Summary paragraph that introduced the three-stage rollout;
  replace with a one-line pointer to the Implementation Specification
  for the rollout plan.
- Strip stage annotations from the code_width_bits table: no more
  "Stage 1 default" / "default, Stages 2+" labels — just clean
  width/footprint/recommendation rows.
- Rewrite the paragraph after the code_width_bits table to drop
  stage framing — "CPU-only deployments adopt the paper's 16-bit
  recommendation; mixed CPU/GPU deployments adopt 12-bit" rather
  than "Stage 1 vs Stage 2."
- Reword "Why this is the right design" to drop "(This is Stage 1
  of the rollout)" parenthetical on the OnPair16-verbatim alternative.
- Reword the Hidden Assumptions GPU-priority bullet to talk about
  "the CUDA decoder, the splits buffer, and the coalesced-format
  flag" rather than "Stages 2 and 3."
- Reword Unresolved Questions #1 (encoder code-width selection
  policy) to drop "Stage 1 ships 16-bit only / Stage 2+" framing.
- Reword Alternatives "Fixed 12-bit format" bullet to drop stage
  framing.
- Drop stage annotations from the Current Vortex state paragraph.
- Add a "Suggested rollout plan" subsection to the Implementation
  Specification with the three-phase content, framed explicitly
  as a project-planning note. The phases are renamed Phase 1/2/3
  to underline they're implementation phases, not design phases.

The actual rollout content survives intact — just relocated and
relabeled. Reviewers reading the design narrative no longer see
staging language threading through the design discussion;
implementers see it in the implementation-spec section where it
belongs.

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

Author: lwwmanning