chore(beads): close f128-restore T1-T6 on 76de18e; file bd-6x4jt0; name the bd-3aktsp flake

T1-T6 closed on one evidence set: the deletion was a contiguous engine, so the
tranches came back together. 53 relit gates run, 52 green; conformance_diff_
c23_narrow went red->green (2 double-rounding divergences fixed).

bd-6x4jt0  log2p1f diverges across the f32 domain (pre-existing, verified RED
           at clean HEAD before and after the restore)
bd-3aktsp  flaking test named: libc_fatal_aborts_child_with_message, 272/0 in
           isolation twice, 271/1 under a 54-target parallel run

bd-ppl81a (T7) remains: 5 f128/C23 targets still dark.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Author: Dicklesworthstone