v2.0.1

Panel disclosure: posterior-backed confidence dots

The default panel now reflects the Bayesian layer that v2.0 turned on, while
staying quiet on the easy cases.

• The Confidence dots are posterior-backed when fusion has run: they show a
  claimed node's posterior support relative to the present/absent decision
  threshold, in three levels (●●● the whole 80% credible interval above the
  threshold; ●●○ the point estimate above but the interval dipping below, thin;
  ●○○ the point estimate below the threshold, the evidence leans against the
  call). The dots are one defined quantity (posterior support); the deterministic
  corroboration stays in the (N sources) text.
• The panel is as confident as its weakest claimed node (a node with fired
  evidence). A declarative node correctly reporting absence does not demote a
  strong verdict.
• When the weakest claimed node is below full confidence, a dimmed line under the
  Confidence row names the claim in plain English: "thin on " or "the
  evidence does not back ".
• --verbose lists each claimed node's posterior and 80% credible interval under
  a labeled heading so the range is not read as a frequentist confidence interval.
• Accessibility: solid versus hollow carries the signal with no color; the hue is
  a second channel; glyphs are limited to ● and ○ for terminal-font safety.
• Without posteriors (--no-fusion) the panel is byte-identical to v1.x, and the
  dot fill is a pure, property-tested function so it cannot drift through the UI.

Deferred past this release: localized dimming of a disputed claim's span in the
Provider line.

Gate: full pytest, ruff, pyright (0 errors), new dot-fill property test and panel
disclosure tests.