subtype: fix sparam extraction for free-typevar singletons
Fix fix_inferred_var_bound to distinguish between typevars that were
peeled from LHS UnionAlls during subtyping (which should be re-wrapped)
and typevars that are inherently free in the LHS (which are singletons
and should be preserved as-is).
When jl_type_intersection_env_s matches a <: b, it now passes the set
of free typevars from `a` (the LHS) to fix_inferred_var_bound. The
function only re-wraps typevars NOT in this set, and returns the
result directly (not wrapped in a TypeVar) when LHS-free typevars
remain. This makes f(Vector.body) correctly bind T = Array{T,1}
(the singleton) rather than T<:Vector (a TypeVar).
For the NamedTuple case (e.g. getindex(NamedTuple{names} where _A)),
the typevar _A from the LHS UnionAll is NOT free in the LHS, so it
gets re-wrapped correctly: T = NamedTuple{names, _A} where _A =
NamedTuple{names} (the original UnionAll).
Also add free-typevar checks in codegen (emit_sparam, uses_specsig,
static_parameter_sym) to treat sparams with free typevars as unknown
at compile time, requiring runtime resolution via the sparam_vals
mechanism.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
