v2.1.2

Security: output-injection sweep

A dedicated sweep of every render path for the class the 2.1.1 render_error fix
revealed: attacker-influenceable strings (CT issuer/subject, federation brand
name, autodiscover domains, per-source error text) reaching the terminal or
markdown without escaping or control-stripping. Most paths were already safe (the
merger scrubs the primary fields and the panel renders via markup-safe
Text.append); four siblings were fixed, each with a regression test:

• render_warning now escapes and control-strips the domain and the per-source
  error reasons (a console.print markup sink carrying server-influenced text).
• render_conflict_annotation (verbose) control-strips the raw candidate values
  it renders, for example a tenant-controlled federation brand name, under
  --explain --verbose.
• The delta command's two error sinks now use the sanitized render_error path
  like every other lookup error.
• The markdown report escapes default_domain and tenant_domains (autodiscover
  values that are control-stripped but not charset-restricted upstream).

The SSRF/network, ReDoS, resource/path/deserialization, and MCP reviews from 2.1.1
stand; no new high or medium issues outside this class.

Gate: full pytest, ruff, pyright (0 errors), validate_fingerprint (841).