Two correctness fixes over the one value-type layout/projection law, plus the
newtype-over-scalar repr the Alloc::alloc proof exposed.

PART A — debt #63 (soundness). A record field initialized from a MULTI-WORD
record VARIABLE dropped its trailing words: `value_repr` classified a
record-bound local by raw word count (`Some(1) => Int`), but a record local is a
ONE-word frame-base POINTER standing for an N-word value, so a 2-word `Region`
field was planned as one `Int` word — under-sizing the frame and letting the
adjacent field clobber the dropped word. Now an aggregate local classifies from
its recorded TYPE first (the full-width `ValueRepr::Record`); scalars/strings
keep the width-count law. One generic fix in `ck_lower::layout::value_repr`.

PART B — debt #38 (named-tuple projection). `plan_tuple` named every slot
`Symbol(0)`, so a named tuple's `layout.field(name)` never matched and `v.name`
died at codegen (receiver span). Now `plan_tuple` carries each component's
label: a positional `Ty::Tuple` still passes `Symbol(0)`, a `Ty::NamedTuple`
passes its label, so projection resolves by the SAME `layout.field(name)` law
records use. The type layer already bridged a positional literal into a
named-tuple slot (assignable.rs) and resolved labels (field_access.rs); this is
the lowering half. `p.0`-positional-on-`Tuple<…>` is left unimplemented: §2a
states the unnamed tuple is "purely positional" but names no positional
projection surface, so no surface was invented.

Newtype-over-scalar repr (§4a). `Alloc::alloc` returns
`(| handle: Handle, arena: Arena |)` where `Handle = newtype usize` — a
newtype-over-a-scalar the layout law could not repr (not a leaf, not a record),
so a Handle-typed tuple/record field failed to plan. Per §4a ("`newtype Celsius
= f64` reaches the machine the same way [as a leaf]"), resolve now records such a
newtype's underlying scalar repr under its own TypeId (no well-known-scalar NAME
added — it stays a distinct nominal), so the existing leaf branch resolves it as
one scalar word.

Proofs (freestanding, exit 42, registered):
- record_field_words — a 2-word `Span` variable placed in a `Box` field,
  crossed by value both ways, both words + trailing scalar read back exact.
- alloc_named_tuple — arena_over → alloc ×2 → project handle (scalar) AND arena
  (record) from each → write/read through both handles, guarded on the advanced
  cursor.
Plus .cktest goldens: named-tuple projection types clean; an absent label
diagnoses E0718 (trapped). arena_return's caveat updated to the fixed truth.

Gates: fmt clean; clippy -D warnings 0; build ok; test full suite green (3
expected canaries); architecture 21/21; conformance ratchet == 94;
stdlib_type_clean green.

@@ -0,0 +93,4 @@
+
+/// `ckc --container <shared> --work <name> run <app> --stdlib <repo>/tests/corpus/stdlib`,
+/// returning the process exit code.
+fn run_ck(name: &str, app: &Path) -> i32 {

@@ -0,0 +14,4 @@
+/// bound was checked against; secondary on the bound's declaration in the
+/// callee's `where`-clause.
+#[must_use]
+pub fn e0710_concrete(