Spec-store bare-name keying: same-named specs from different modules silently clobber

[hierophantos]

Found by the Numerics N6d-i grounding-audit (2026-07-02, HEAD c8a425f).

The spec registry keys by BARE symbol with silent last-write-wins:

• register-spec! (macros.rkt:480-482)
• import spec-propagation (driver.rkt:2810-2811)
• implicit-hole counting strips FQNs before lookup (elaborator.rkt:567-576)

Two same-named specs from different modules (e.g. nat's spec add Nat Nat -> Nat vs a generic spec add {A} ... where (Add A); or any two libs both defining reduce) overwrite each other in any module importing both. The loser's call sites get WRONG implicit-argument counts — silently (no diagnostic; wrong elaboration downstream).

Fix direction: FQN-keyed spec store, or module-scoped shadowing with deliberate resolution order. Crosses the module system — PM-series candidate.

Blocks: lifting the add/sub/mult method-wrapper derive skips (DEFERRED.md § "Numerics N6d-i follow-ups" item 1).

[hierophantos]

Numerics N6e update (2026-07-02): the plan of record supersedes lifting the add/sub skips -- N6e-E2 binds + - * / negate as trait-backed VALUES (fresh names, no spec-clobber; design doc section 9d D-N6.6 mini-design), which delivers the ergonomic first-class operators without touching the add/sub derive skips. The skip set (add sub join reduce) stays; capture is already structurally fixed (resolution reorder); spec-clobber remains the only reason and is tracked by DEFERRED.md section N6d-i item 2.