annotate a republished binding as typeof the binding it republishes, which ZodType widened into a lie

[eburgos]

An item that publishes another item's binding — an alias's const, a one-slot
tuple struct over a brand, and the generic $SchemaDefault of both — annotated
it ZodType<T>. Over a branded schema that annotation does not hold:
$ZodBranded's output is the branded type while ZodType<T>'s parameter is
the unbranded one, and tsc --strict refuses the assignment. Thirteen such
errors on one emitted module, across every brand inner and in both declaration
orders.

The distinction the fix draws is structural, not nominal: an item that builds
its own expression keeps ZodType<Name>, and an item that restates a sibling's
binding says so — typeof {Name}$RawSchema, the one annotation that is true
whatever the sibling's class is. The generic default binds its factory call to
a raw const first so the same typeof applies. That is why the registry answer
was rejected on measurement: an alias written above its brand has no entry to
read, and an alias chain re-breaks at the second link — one macro invocation
sees one item, and typeof needs to know nothing about the sibling but its
name. The ReturnType<typeof Factory> spelling type-checks while silently
collapsing an inferred value to the opaque parameter, which is the trap
assignability alone does not catch.

Nothing that builds its own expression moves: an alias of a String, of a
Vec, of a map, every composite over a brand, and every brand's own binding
are pinned byte-identical, and no pre-existing expectation string changed.
Measured on the crate's real emission: thirteen TS2322s before, zero after,
declaration emit clean, and the inferred type of every parse stays exact.

just fmt, just check, just quick, just lint-all across all 68 toggles, and the
test powerset across all 68 in four partitions — all green.