oracle: an ill-conditioned quantity gets a wider fixture band; Scordelis-Lo re-blessed

Main's oracle job failed at 8cba3a7: scordelis-lo-roof-solid u_z moved from
-0.3014029226397275 (blessed before #50) beyond its 1e-9 band. The case is a
thin shell (R/t = 100) of one layer of hex20R solids, ill-conditioned, so
rounding alone moves u_z by about 1e-9 relative: on macOS #50's multithreaded
supernodal factorisation moved it by 8.2e-10 (to -0.3014029223929), #52's
interpolation order by a further 1.3e-10 (to -0.3014029223541), and Linux
CI by slightly more, past the band. #52 kept the lattice geometry the same.
The result itself is unchanged at the 1e-9 level (0.33% from the
reference).

A quantity may now set `fixture_relative` (with a required
`fixture_reason`) for a wider fixture band; Scordelis-Lo uses 1e-7 and is
re-blessed. Moved-fixture messages print full precision.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NmAttHhUZTokgPeEJia8sm

docs/verification.md

@@ -64,7 +64,8 @@ cargo xtask oracle --quick              # skip cases marked "slow"
   case cites its reference values (or `null` until transcribed) and has no recipe yet.
 - `fixture.json` holds each passing quantity's accepted `value` and its `band`
   (10⁻⁹ of the larger of value and reference, plus 10⁻³ of an absolute tolerance; a quantity whose
-  reference is zero, which is rounding noise, keeps its absolute tolerance as its band): the result is
+  reference is zero, which is rounding noise, keeps its absolute tolerance as its band; an
+  ill-conditioned quantity may set a wider `fixture_relative` with a `fixture_reason`): the result is
   deterministic, the band only absorbs last-digit differences of the system maths library.
 
 The run fails when a quantity misses its reference (unless it is a `known_failure`, which is