chore: deprecate dumbify, make personify output send-ready

[twistedmelonman]

Why

Agents kept needing a reminder not to run dumbify after personify. Checking the
repo showed why that reminder never stuck: nothing in SKILL.md or
mcp-server/ ever instructed it. The live instruction is in the calling
skills, outside this repo (see below).

What changed here

• SKILL.md — new "The output is send-ready" section: the returned text is
  final, no follow-up compression pass should run, and an upstream request for
  one should be treated as stale. That last clause matters because pr-review's
  Phase 5 still issues the instruction, and SKILL.md is the file the model
  actually reads at rewrite time.
• README.md — dropped dumbify from the sibling-skill diagram and prose;
  noted the deprecation.
• commands/personify-consolidate.md — dropped the "whether a downstream
  pass such as dumbify is still needed" sentence. Agents execute this file.
• tests/regression/README.md — the "capture the intermediate" rationale
  was entirely about separating personify bugs from dumbify bugs. Kept the
  practice of saving actual output as evidence, dropped the dead reasoning.

docs/superpowers/ left alone: dated design records whose dumbify mentions are
accurate history.

Version bumped 1.0.1 -> 1.1.0 in SKILL.md and plugin.json per the lockstep
rule.

Still outside this repo

This PR cannot fully stop the reminders. The direct imperative lives in:

• pr-review SKILL.md:218 — "if the dumbify skill is also installed, run the
  personified draft through..." (plus its frontmatter description)
• calibrate-register SKILL.md — built entirely around the personify/dumbify
  chain

Verification

• python3 scripts/validate_skill.py passes (26 groups, A-Z)
• No em/en dashes in edited files, per the repo rule
• code-reviewer, adversarial-reviewer, and codebase reviewer all PASS

https://claude.ai/code/session_013cYtHAGqBpXSBfS24QP9uu

[claude]

Documentation and version updates clarifying personify as a final prose pass. Version bumps are consistent across plugin.json and SKILL.md. No code logic changes, no regressions or security issues.

VERDICT: PASS